Files
aimsight-labeling/docs/archive/label-map.md
T

115 KiB

Label map

  • For the labelling agent, the analyst and the reviewers: the labeled phenomena, sorted by phenomenon and not by station; what is seen, how the cause was known, what is concluded from it, and the labels that show it.
  • Sources, as of 2026-11-05: the 23 reviewed labels of the expert (labeling/manual/, 6 stations) and his words take precedence; the 991 reviewed automatic labels (labeling/auto/, 11 stations); the 13 events (events/), each attached to its station by its data; the cases of labeling/knowledge/. Key of a label: STATION DD.MM HH:MM (its start, UTC), with the sensor or the family in brackets when two labels share a start; the same key may designate a label of the expert and an automatic one redone at the same instant. [E-xxx]: an event.
  • Derived: redone in full by /labels-map; it is not corrected by hand: a correction is made in the label.
  • The other map, events/carte-des-evenements.md, lists what can happen, seen or not; this one only says what has been labeled. Qualitative: quantifying is the job of /labels-learner.

Table of contents

Read the table of contents and sections 2 to 4 in full, then the card of each phenomenon met (Read with offset at its line).

  • A. No measurement
    • A gap common to several stations (line 71): the measurements are missing at the same minutes on all the stations that measure at that moment;
    • The station runs, the measurement is missing (line 81): the station is powered and communicates, but a measurement is not made or not recorded;
    • The station goes silent on its own, for a long time (line 91): the whole station stops on its own, without the signature of a lack of energy: no measurements, no images, no metrics, for hours or days.
    • The station shuts down for lack of energy (line 100): the battery does not last the night: the station shuts down, then restarts with the solar charge.
    • Mains cut: the station runs on its battery (line 110): a station on mains loses its power: it stops, or restarts on its battery without charge, until the mains return.
  • B. The view is obstructed
    • A pass-by hides the targets for the length of one measurement (line 121): something passes between the camera and the targets (a train, a vehicle, passers-by) and entirely hides part of them, for a single acquisition.
    • A target left under the threshold latches onto a headlight, a reflection or another spot (line 135): a pass-by brings a bright point near a target;
    • A vehicle or machine that stays in front of the targets (line 145): a parked vehicle, a machine or an object stays in front of one or a few targets, sometimes a reference, for hours or days;
    • An animal very close to the lens (line 158): an animal passes very close to the lens (the spider itself, flying insects, perhaps a bird);
    • A spider web in front of the lens (line 168): a web stretched very close to the lens veils part of the scene;
    • A plant shaken by the wind in front of a target (line 182): a foreground plant, pushed by the wind, passes at times in front of a target.
    • An obstacle covers part of the spot (line 191): an obstacle, or a mark on the glass, hides part of the spot of a target, sometimes of a few, for hours or weeks;
    • The soiled glass (line 201): after rain, drop marks remain on the glass in front of the lens and veil part of the scene, durably.
    • Rain (line 210): drops in front of the lens veil the spots during the hours of rain, at night above all [NIEDERHORN 01.10 04:00, NIEDERHORN 02.10 02:00], by day too [JUNGFRAU 24.10 21:00].
    • Visibility drops over the whole scene, without an obstacle (line 222): the whole scene is seen less well, with nothing in front of the targets [JUNGFRAU 07.10 08:36].
    • Part of the scene goes dark at night (line 239): at night, only part of the scene loses its light, without an obstacle: the distant spots [LENZSPITZE 16.10 22:25], or a few neighbouring targets at the edge of the group [NORDEND 30.09 20:45].
    • A reference loses its brightness, without an obstacle (line 249): the spot of one or two references weakens and shrinks with nothing in front, most often at night, then comes back on its own;
    • Atmospheric turbulence (line 262): layers of air at different temperatures disturb the image;
    • The sun in the field or behind a spot (line 272): the light of the sun drowns a spot: the sun enters the field [NIEDERHORN 31.10 11:00], or it lights the background of a target or a reference, as bright as it is [NORDEND 12.08 13:30, JUNGFRAU 12.10 15:45].
    • A faint spot lost when the light is missing (line 282): a target whose spot is already faint goes above its threshold when the light is missing: transition from night to day [FLORIDA 01.10 07:15], overcast sky [NORDEND 16.08 09:15], noisy night image [NORDEND 08.09 01:45].
  • C. The target has changed
    • A target rejected at all its measurements (line 293): a target is never accepted, whatever the weather;
    • A target right at the threshold, the whole batch (line 302): a target lives permanently at the limit of detection: its error_value keeps crossing the threshold.
    • A target rejected every night or every day (line 311): a target goes above its threshold under the ordinary regime of the station (floodlight at night, daylight), at the same hours, whatever the weather.
    • A target that degrades, and sometimes recovers (line 320): a target that used to be seen well is seen worse and worse from a date on, and does not come back, or only after weeks.
    • The configuration of a sensor has changed (line 330): a sensor is reconfigured, or a setting changes (sensor moved, starting position changed);
  • D. The references
    • A reference dropped: the base changes (line 341): a reference is seen badly, its error_value exceeds the threshold, the system drops it.
    • Two references lost: the camera is blind (line 355): two of the three references become almost invisible, and the system drops them one after the other;
    • A compensation on two references (line 365): the compensation runs on only two references, the station having only two or having lost one for good;
    • The base changes without a reference lost (line 375): all the references stay in the compensation, but the distance between them changes a little;
  • E. The camera and its mount
    • The camera moves through a thermal effect (line 386): the lens or the mount of the camera expands and contracts with the temperature;
    • The camera drifts away when the temperature of its housing drops, without a daily cycle (line 406): sheltered, in a building or a tunnel, the camera has no daily cycle;
    • A step of the camera that the compensation does not erase (line 416): the camera moves all at once;
    • The camera tilts little by little (line 425): the camera nods forward little by little, over weeks, and the compensation of this movement is incomplete: it corrects some targets better than others [BREITHORN 03.10 00:00, according to the expert];
  • F. The structure has moved
    • The structure expands and contracts with the temperature (line 437): a structure that heats up and cools down moves, normally for its kind;
    • A rise over several days with the drop in temperature (line 447): the temperature keeps dropping for days, which breaks the thermal cycles;
    • A real movement of the structure (line 456): the structure really moves, and drags the camera and some references along.
    • Targets move slowly, alone, then come back (line 465): one or a few targets change position slowly, alone, for hours or days, without the camera moving or the view being obstructed, then come back, partly or beyond;
    • A target or a small group jumps and stays shifted (line 475): a target, or a few neighbours, move all at once to another level and stay there, without the camera moving or a cause being visible;
    • A step taken during a measurement outage (line 485): during a stop of the measurements, targets change level;
  • G. Station configuration
    • Targets that are twins of their references (line 496): the same prism serves as target and as reference, for lack of a possible reference in this part of the field, because the system does not yet accept a sensor with two roles.
  • 2. Reading rules (line 505)
  • 3. False friends (line 538)
  • 4. The context of the stations (line 559)
  • 5. Differences with the event map (line 578)
  • 6. What the labels do not say yet (line 596)

1. The phenomena

A. No measurement

A gap common to several stations

  • What happens: the measurements are missing at the same minutes on all the stations that measure at that moment; the cause is common and outside the stations (transmission or reception of the data), not a cut or a failure of each one; it is not established.
  • What is seen: no measurement for any sensor, often without an image; the voltage is normal around the gap; the metrics are often missing at the same minutes [CLARIDEN 26.10 14:30], including the hourly metrics of the slow-cadence stations [CLARIDEN 19.10 20:45]. A station may run during the gap: its unsent entries pile up, leave at the end of the morning, and its measurements of the night are nevertheless missing [JUNGFRAU 15.10 22:45]. The curves resume at the previous level [BREITHORN 15.10 20:00, NEBRASKA 15.10 20:00].
  • Over time: from one measurement to one hour, by day as by night. The night of 15 to 16.10, all the stations: the gap starts around 23:00 everywhere, the resumption varies by station, most at the end of the morning [JUNGFRAU 15.10 22:45, NIEDERHORN 15.10 22:00 (No data)].
  • The cause, and how it was known: deduced from the coincidence between stations and from the station that runs during the gap [JUNGFRAU 15.10 22:45]; with only two stations, "possible" [CLARIDEN 20.10 06:10, NIEDERHORN 20.10 20:00]. The three stations of the same building lose their data that night like the others: the cause is not the building [MICHIGAN 15.10 20:00].
  • What is concluded: the curves are missing; nothing to do on the station (for the night of 15 to 16, "if it does not repeat").
  • Variants: a one-measurement gap at night, around 02:00, on a pair of stations only, several nights [FLORIDA 09.10 01:45, SNCF01 09.10 01:45]; other pairs in other weeks [NORDEND 05.09 01:30, WILDSTRUBEL 05.09 02:00]; probable common cause, nothing to do if it does not get worse. The power is missing as well (see "Mains cut") [CLARIDEN 15.10 22:45]. On return, targets resume at another level (see "A step taken during a measurement outage") [NIEDERHORN 15.10 22:00 (No data)]. A gap dated by the data documentation, the other stations being missing from the folder [NORDEND 19.08 14:15]; it may open a hindered sending, the pending entries piling up until the next morning [NORDEND 22.08 16:30].
  • Labels: JUNGFRAU 15.10 22:45, MICHIGAN 15.10 20:00, CLARIDEN 26.10 14:30, LENZSPITZE 19.10 09:05, FLORIDA 09.10 01:45, NORDEND 05.09 01:30; and 67 others: NORDEND 15, SNCF01 12, FLORIDA 10, CLARIDEN 8, LENZSPITZE 8, JUNGFRAU 5, NIEDERHORN 5, WILDSTRUBEL 2, BREITHORN 1, NEBRASKA 1.

The station runs, the measurement is missing

  • What happens: the station is powered and communicates, but a measurement is not made or not recorded; it alone is affected.
  • What is seen: a measurement is missing for all the sensors; at that instant the station synchronizes, or the image exists, and the voltage is normal [JUNGFRAU 05.10 06:00, NIEDERHORN 08.10 19:00, NORDEND 15.08 01:00]; sometimes the image is missing too, the synchronization is there [JUNGFRAU 02.10 11:15]. Resumption without a jump.
  • Over time: one measurement, rarely two or three [NORDEND 20.08 08:00]; several times the same day [JUNGFRAU 02.10 11:15, JUNGFRAU 02.10 12:00, JUNGFRAU 02.10 13:15], or around the same hour on many nights [NORDEND 19.08 01:15].
  • The cause, and how it was known: not established; that the station runs is established by the metrics, the synchronization or the image; it is still necessary that no other station has the gap [NORDEND 19.08 01:15].
  • What is concluded: nothing to do.
  • Variants: a few hours without measurement or image, metrics and synchronizations continuing: the acquisition stops, the station communicates; to check if it repeats [NORDEND 28.08 07:45].
  • Labels: JUNGFRAU 05.10 06:00, NIEDERHORN 08.10 19:00, NORDEND 28.08 07:45, CLARIDEN 03.10 07:40, NORDEND 07.09 01:00; and 27 others: NORDEND 11, JUNGFRAU 8, NIEDERHORN 7, CLARIDEN 1.

The station goes silent on its own, for a long time

  • What happens: the whole station stops on its own, without the signature of a lack of energy: no measurements, no images, no metrics, for hours or days.
  • What is seen: everything is missing, on this station only; the voltage is normal just before and after [LENZSPITZE 23.10 23:55, CLARIDEN 03.10 17:45], or the battery was not low [WILDSTRUBEL 09.10 12:30]; without power metrics, nothing tells the cause [SNCF01 22.10 15:00].
  • Over time: from a few hours to more than two weeks; between the cuts, isolated measurements [SNCF01 20.10 12:45, SNCF01 22.10 10:30]; metrics and images may come back before the measurements [WILDSTRUBEL 09.10 12:30]. A gap lasting to the end of the batch is not an end-of-file delay when nothing arrives in the following days while the other stations have their data [WILDSTRUBEL 31.10 20:15].
  • The cause, and how it was known: not established (acquisition, transmission, retrieval; station stopped). On resumption, the camera has moved and a blurred object is very close to the lens: an intervention is possible [CLARIDEN 03.10 17:45].
  • What is concluded: if it repeats, check the transmission [LENZSPITZE 23.10 23:55], the power [SNCF01 22.10 15:00], or contact the operator [WILDSTRUBEL 09.10 12:30].
  • Labels: CLARIDEN 03.10 17:45, LENZSPITZE 23.10 23:55, SNCF01 22.10 15:00, WILDSTRUBEL 09.10 12:30, WILDSTRUBEL 31.10 20:15, SNCF01 20.10 12:45; and 4 others: SNCF01 3, WILDSTRUBEL 1.

The station shuts down for lack of energy

  • What happens: the battery does not last the night: the station shuts down, then restarts with the solar charge.
  • What is seen: no measurement, the metrics stop with them; just before, the battery at its lowest, the voltage under its usual night level [WILDSTRUBEL 01.10 03:15]; a battery that stops dropping during the gap [WILDSTRUBEL 02.10 20:15]. In the early morning, a few isolated measurements, with rejected targets [WILDSTRUBEL 06.10 19:45], or almost all of them, without realignment [WILDSTRUBEL 03.10 15:30]; on resumption, the computing board still cold [WILDSTRUBEL 01.10 03:15].
  • Over time: every night, from a few hours at the end of the night [WILDSTRUBEL 29.09 02:30] to the whole night, sometimes from the afternoon [WILDSTRUBEL 07.10 13:30]; resumption at sunrise.
  • The cause, and how it was known: deduced from the metrics (battery and voltage before the gap, resumption with the charge).
  • What is concluded: the structure does not move; the station lacks energy: check battery and panel.
  • Variants: an outage of nearly two weeks starts the same way; the station restarts with a full battery, a voltage much higher than usual, as on an external power supply, and the tilt of the camera no longer has its previous level: check battery, panel, possible intervention [WILDSTRUBEL 09.09 21:45].
  • Labels: WILDSTRUBEL 01.10 03:15, WILDSTRUBEL 09.09 21:45, WILDSTRUBEL 03.10 15:30, WILDSTRUBEL 07.10 13:30, WILDSTRUBEL 29.09 02:30, WILDSTRUBEL 06.10 19:45; and 6 others: WILDSTRUBEL 6 (one station).

Mains cut: the station runs on its battery

  • What happens: a station on mains loses its power: it stops, or restarts on its battery without charge, until the mains return.
  • What is seen: during the gap, no measurements, no images, no metrics; on resumption, the battery drawn down and the voltage well under its usual level, without charge, for hours [CLARIDEN 18.10 13:45]; then the charge resumes [CLARIDEN 15.10 22:45].
  • Over time: about ten hours without data [CLARIDEN 18.10 13:45]; or a night, within the gap common to all the stations, then the battery alone until early afternoon [CLARIDEN 15.10 22:45].
  • The cause, and how it was known: deduced from the metrics, against the usual level of the station, even without going under the knee of the battery.
  • What is concluded: if it repeats, check the power supply of the site [CLARIDEN 15.10 22:45].
  • Labels: CLARIDEN 18.10 13:45, CLARIDEN 15.10 22:45 (one station).

B. The view is obstructed

A pass-by hides the targets for the length of one measurement

  • What happens: something passes between the camera and the targets (a train, a vehicle, passers-by) and entirely hides part of them, for a single acquisition.
  • What is seen:
    • targets: the error_value of the hidden targets rises almost to 1, their measurements are rejected and removed from the display; a few targets left under the threshold keep a small jump [NIEDERHORN 30.09 21:00, NIEDERHORN 01.10 11:00], or a false point (see "A target left under the threshold latches onto…") [LENZSPITZE 17.10 05:00];
    • references: intact when the pass-by is only in front of the targets [NIEDERHORN 30.09 21:00, NIEDERHORN 01.10 11:00]; affected when it passes in front of them: a single one dropped and the whole base jumps [JUNGFRAU 06.10 05:45, BREITHORN 07.10 08:00]; half hidden without being dropped, it also distorts the base [BREITHORN 10.10 00:00]; none kept, and the camera is not realigned: the points of the instant have no meaning [WILDSTRUBEL 03.09 08:00, WILDSTRUBEL 15.10 07:45, SNCF01 05.10 15:45];
    • image quality, on a recent camera: the flux of the hidden targets falls to almost nothing, the others keep their light; the list of cut fluxes is that of the hidden targets [LENZSPITZE 17.10 05:00, LENZSPITZE 16.10 07:30];
    • images: a train, motion blur, wagon numbers readable [NIEDERHORN 01.10 20:00, NIEDERHORN 01.10 11:00]; at dawn, a dark band with a straight right edge across the scene [NIEDERHORN 06.10 11:00]; at night, big bright spots [SNCF01 30.10 17:45], headlights, a plate [WILDSTRUBEL 03.09 21:30]; a vehicle or passers-by in transparency, in blurred copies [WILDSTRUBEL 07.10 11:15, WILDSTRUBEL 07.09 14:00]. Often nothing: no image at the instant, or an image of the same minute where the pass-by is not there [LENZSPITZE 17.10 05:00, SNCF01 07.10 19:30].
  • Over time: one measurement, sometimes two [WILDSTRUBEL 07.09 14:00, SNCF01 30.10 21:15]; everything comes back at the next one; by day as by night, often several times a day. A target still rejected alone at the following measurement is the end of the same pass-by [WILDSTRUBEL 07.10 11:45].
  • The cause, and how it was known: seen on the image of the hour [NIEDERHORN 30.09 21:00, NIEDERHORN 01.10 11:00]; otherwise deduced from the signature, the same as that of the pass-bys seen [SNCF01 07.10 19:30, LENZSPITZE 17.10 05:00]: two neighbouring references lost together for a single measurement are enough [WILDSTRUBEL 28.09 05:15, WILDSTRUBEL 30.09 10:45]; a single target sometimes loses its spot without what hides it being visible [FLORIDA 20.10 05:15].
  • What is concluded: the structure does not move, nothing to do; the point is to be ignored. "The tracks do not move" [NIEDERHORN 30.09 21:00].
  • Variants: a train on the back track hides only the targets on the left, sometimes just one [NIEDERHORN 03.10 09:00, NIEDERHORN 26.10 21:00]; a partial pass-by, a group of low targets [SNCF01 06.10 06:30], or two targets just above the threshold, cause not established [SNCF01 12.10 06:30]; very close to the camera, all the references and almost all the targets rejected [WILDSTRUBEL 03.09 08:00, WILDSTRUBEL 06.10 10:15]; at sunrise, references already dropped by the light, one passer-by is enough to leave none [WILDSTRUBEL 06.09 05:45]. Without hiding anything: the flux stays whole, the error_value of a group rises and a second bright pattern appears, a pass-by next to or behind the targets, or its reflection; nothing on the images, which are only at the full hours, according to the expert [LENZSPITZE 31.10 10:50, LENZSPITZE 27.10 10:50].
  • Labels: NIEDERHORN 30.09 21:00, NIEDERHORN 01.10 11:00, JUNGFRAU 06.10 05:45; NIEDERHORN 01.10 20:00, SNCF01 05.10 15:45, WILDSTRUBEL 01.09 10:45, WILDSTRUBEL 07.09 14:00, LENZSPITZE 17.10 05:00, BREITHORN 07.10 08:00; and 135 others: WILDSTRUBEL 73, NIEDERHORN 22, LENZSPITZE 18, SNCF01 17, BREITHORN 3, CLARIDEN 1, FLORIDA 1; [E-005], the trains of NIEDERHORN 30.09 21:00 and 01.10 11:00.

A target left under the threshold latches onto a headlight, a reflection or another spot

  • What happens: a pass-by brings a bright point near a target; the measurement latches onto it while staying under the threshold, and the point kept is false.
  • What is seen: a single target makes a very large jump of corr_x and corr_y for the length of one measurement, its raw position jumps [SNCF01 17.10 09:45, SNCF01 16.10 11:30]; its neighbours rejected at the same instant [WILDSTRUBEL 04.09 06:30], or nothing else moves [WILDSTRUBEL 01.09 14:30, WILDSTRUBEL 29.09 21:00]. Images: bright points near the target, absent at the previous measurement and at the next one [WILDSTRUBEL 01.09 14:30, WILDSTRUBEL 06.09 14:15]: the lights, the reflectors or the bodywork of a vehicle; the sun on the roadway, for its part, does not change from one measurement to the next [WILDSTRUBEL 28.09 05:45].
  • Over time: one measurement.
  • The cause, and how it was known: seen on the thumbnail; deduced without an image, like a pass-by of the same kind seen another day [SNCF01 28.10 10:15].
  • What is concluded: the structure does not move; the point is false and stays in the corrected curve: only a rejected measurement is removed from it [NIEDERHORN 30.09 21:00].
  • Variants: a reference latches onto a reflection in the same way, is rejected and the base changes [WILDSTRUBEL 23.09 09:15]; a parked vehicle has the same effect [WILDSTRUBEL 05.09 03:30].
  • Labels: WILDSTRUBEL 01.09 14:30, WILDSTRUBEL 29.09 21:00, WILDSTRUBEL 04.09 06:30, SNCF01 17.10 09:45, SNCF01 16.10 11:30; and 8 others: WILDSTRUBEL 7, SNCF01 1.

A vehicle or machine that stays in front of the targets

  • What happens: a parked vehicle, a machine or an object stays in front of one or a few targets, sometimes a reference, for hours or days; always the same sensors.
  • What is seen:
    • targets: the same sensors rejected at every measurement, their curves absent; return without offset when the vehicle leaves [BREITHORN 03.10 00:00]; a partly hidden target jumps, or slides over the visible part of its spot [BREITHORN 23.10 16:00];
    • references: they most often hold; a hidden reference leaves the compensation on the others, without a jump or barely, on a station that has more than three [BREITHORN 03.10 00:00, BREITHORN 26.10 20:00]; with two remaining, the base changes, a very bad sign [BREITHORN 25.10 20:00]; with none, no realignment and a target still measured makes a peak [WILDSTRUBEL 05.09 03:30, WILDSTRUBEL 26.09 03:45 (Obstruction)]; hidden without being dropped, all the targets jump [BREITHORN 04.10 12:00, LENZSPITZE 30.10 13:50]. A hidden reference above its threshold keeps in the export corrections of tens of pixels, pointed at the car: it is not a movement [E-012, WILDSTRUBEL 26.09 03:45 (Obstruction), seen in the data];
    • images: the vehicle (a cab, a grille, headlights, an oblique structure) [BREITHORN 06.10 16:00, BREITHORN 25.10 20:00]; at night, its light veils the image [BREITHORN 08.10 20:00] and its plate shines under the infrared [WILDSTRUBEL 05.09 23:00]; a lit machine darkens the whole image [NIEDERHORN 13.10 04:00]; the arm of a machine deployed, folded back before and after [LENZSPITZE 30.10 13:50]; on a recent camera, the light of a few targets cut, of a nature not seen [LENZSPITZE 17.10 10:00].
  • Over time: from one or two measurements (a machine) [NIEDERHORN 13.10 04:00] to a few hours (cars parked at sunrise) [WILDSTRUBEL 26.09 03:45 (Obstruction), NIEDERHORN 24.10 16:00], a night [BREITHORN 06.10 16:00], nearly two days [BREITHORN 03.10 00:00]; night after night at the same place [BREITHORN 21.10 16:00].
  • The cause, and how it was known: seen on the images; deduced, when nothing is visible, from the signature of the nights when a vehicle is parked: probable, not established [BREITHORN 22.10 16:00]; established by the light being cut [LENZSPITZE 17.10 10:00].
  • What is concluded: the structure does not move: ignore these sensors over the range; with two references remaining, do not trust the data of the range [BREITHORN 25.10 20:00].
  • Variants: something in front of the targets for a quarter of an hour, not seen, without rejection, with large jumps of corr_x [LENZSPITZE 27.10 17:35]. An intervention at the camera: a large blurred object against the lens, a worker; the scene shifts and the camera tilts, then comes back [CLARIDEN 05.10 10:30].
  • Labels: BREITHORN 06.10 16:00, BREITHORN 25.10 20:00, WILDSTRUBEL 26.09 03:45 (Obstruction), NIEDERHORN 24.10 16:00, LENZSPITZE 30.10 13:50, CLARIDEN 05.10 10:30; and 17 others: BREITHORN 11, WILDSTRUBEL 3, LENZSPITZE 2, NIEDERHORN 1; [E-012] (learned, not revalidated), the morning of WILDSTRUBEL 26.09 03:45.

An animal very close to the lens

  • What happens: an animal passes very close to the lens (the spider itself, flying insects, perhaps a bird); lit at night, it hides part or all of the scene, for one or two measurements.
  • What is seen: the error_value of the references and of most of the targets jumps almost to 1, then comes straight back [NORDEND 31.07 20:15]; almost all the targets rejected, the others jump [NORDEND 27.08 21:00, NIEDERHORN 13.10 22:00]; a single reference rejected and all the targets jump together [NORDEND 18.08 17:45, NIEDERHORN 09.10 15:00]; or the references intact [NIEDERHORN 29.10 20:00]; or a single target that drifts away for an hour, the error_value of almost all of them a little higher, without rejection [NORDEND 10.08 01:15]. Images: the spider, its body lit by the infrared, as a big white blurred spot (image of 20:30) [NORDEND 31.07 20:15, NORDEND 07.10 19:30]; insects as large white blurred spots under the floodlight [NIEDERHORN 13.10 22:00]; a dark mass in front of a reference [NIEDERHORN 09.10 15:00].
  • Over time: one measurement, sometimes two, an hour [NORDEND 10.08 01:15].
  • The cause, and how it was known: seen on the image [NORDEND 31.07 20:15]; its nature sometimes not established: spider or insect [NORDEND 10.08 01:15], bird or insect [NIEDERHORN 09.10 15:00].
  • What is concluded: the structure does not move, nothing to do; the point is to be ignored.
  • Variants: after the spider, a new web stays and veils, more faintly, a reference and a few targets (see the web) [NORDEND 31.07 20:15].
  • Labels: NORDEND 31.07 20:15; NORDEND 18.08 17:45, NORDEND 07.10 19:30, NORDEND 10.08 01:15, NIEDERHORN 13.10 22:00, NIEDERHORN 09.10 15:00, NIEDERHORN 29.10 20:00, and 1 other: NORDEND 1.

A spider web in front of the lens

  • What happens: a web stretched very close to the lens veils part of the scene; labeled Visibility when it veils the image, Obstruction when a thread hides only a few spots.
  • What is seen: depending on its place and its hold:
    • the references in turn: their error_value rises clearly above its previous level, many measurements of the targets are rejected; the compensation drops one or two references, and the corr_y of many targets jumps in steps (see "A reference dropped") [NORDEND 28.07 20:45];
    • without any signal: at night, the web saturates in white like the targets and entirely hides CTR04; its error_value barely rises, the reprojection does not move, "no measurement is rejected"; the system believes it sees three references when it has only two [NORDEND 03.09 18:01]; later the same evening, a thread on PR07 makes it reject a few measurements [NORDEND 03.09 22:45];
    • a few targets: their error_value rises almost to 1, rejected at times, and a slightly disturbed reference makes a good part of the targets move slightly [NORDEND 04.09 16:09]; targets rejected all night [NORDEND 18.08 17:30]; at dusk, most rejected together [NORDEND 21.08 15:45];
    • a thread on a spot: the covered target rejected at times, its neighbours measured [NORDEND 24.08 11:00, NORDEND 24.09 11:15]; laid on a reference, it makes it dropped: all the targets on a step while it is missing, small rebound on its return [NORDEND 18.08 22:15]; or without effect on the curves [NORDEND 03.09 22:45];
    • images: the web as broad blurred bands, at night as by day; the image becomes sharp again when it leaves [NORDEND 28.07 20:45]; a thread as a light band under the infrared, invisible by day when the spot stays faint [NORDEND 24.08 11:00], or lit by the morning sun [NORDEND 26.08 07:00].
  • Over time: from an hour to a little over two days; it moves, then leaves on its own; often woven at the end of the day, absent the next day [NORDEND 18.08 17:30, NORDEND 07.10 17:00]; the same spots may stay out when it is no longer seen [NORDEND 07.10 17:00].
  • The cause, and how it was known: seen on the images; deduced when threads cross the field those nights without being seen on the target [NORDEND 16.09 01:30, NORDEND 17.09 20:00]; a blurred band that can hardly be guessed, cause then not established [NORDEND 28.08 15:15].
  • What is concluded: the structure does not move, nothing to do; ignore the covered targets, and all the curves when a reference is dropped [NORDEND 18.08 22:15]. The case without signal is probably a flaw of the system: it can silently distort values, only the quality metrics would see it [NORDEND 03.09 18:01].
  • Labels: NORDEND 28.07 20:45, NORDEND 03.09 18:01, NORDEND 04.09 16:09; NORDEND 18.08 22:15, NORDEND 24.08 11:00, NORDEND 24.09 11:15, NORDEND 21.08 15:45, NORDEND 18.08 17:30, NORDEND 07.10 17:00, and 9 others: NORDEND 9 (one station); [E-002], the web of NORDEND 28.07 20:45, taken up again on 31.07 in the evening.

A plant shaken by the wind in front of a target

  • What happens: a foreground plant, pushed by the wind, passes at times in front of a target.
  • What is seen: this target alone: its error_value jumps about one measurement in two, then comes back [NIEDERHORN 02.10 09:00]; rejected at times, or, just under the threshold, a one-measurement peak of corr_x or corr_y; some days, peaks without any rejection [NIEDERHORN 18.10 06:00]. Its neighbour is not affected [NIEDERHORN 02.10 09:00], or is too, rejected for one measurement [NIEDERHORN 08.10 16:00]. Images: at the hours of the peaks, the leaves cover the target, its spot pales or disappears [NIEDERHORN 05.10 13:00], or looks cut [NIEDERHORN 20.10 05:00]; at the other hours, they are beside it.
  • Over time: from a few hours to almost a day, almost every day of the month, by day as by night [NIEDERHORN 02.10 07:00, NIEDERHORN 03.10 06:00].
  • The cause, and how it was known: seen on the images, hour by hour [NIEDERHORN 02.10 09:00].
  • What is concluded: the structure does not move, nothing to do; ignore the curve of this target over the range.
  • Labels: NIEDERHORN 02.10 09:00; NIEDERHORN 05.10 13:00, NIEDERHORN 08.10 16:00, NIEDERHORN 18.10 06:00, NIEDERHORN 02.10 07:00, NIEDERHORN 03.10 06:00, NIEDERHORN 23.10 11:00, and 17 others: NIEDERHORN 17 (one station); [E-005], the plant of NIEDERHORN 02.10 09:00.

An obstacle covers part of the spot

  • What happens: an obstacle, or a mark on the glass, hides part of the spot of a target, sometimes of a few, for hours or weeks; the measured centre moves.
  • What is seen: the target drifts slowly and steadily away from its neighbours, which do not move, its error_value rising with the slide without reaching the threshold [CLARIDEN 18.09 00:00]; or it changes level in one step, alone, and stays there, error_value higher [WILDSTRUBEL 24.09 02:30, NORDEND 08.10 02:15]; it may exceed the threshold most of the day [WILDSTRUBEL 03.09 02:45]; on a recent camera, its flux drops in part and its corr_x jumps, without rejection [LENZSPITZE 27.10 22:00]. A reference partly covered, rejected without interruption, makes the base change [WILDSTRUBEL 02.10 12:45]. Images: the spot cut on one side by a dark band, when that of its neighbour of the same model is whole [CLARIDEN 18.09 00:00]; smaller at night than the other nights [WILDSTRUBEL 24.09 02:30]; a blurred light disc on the glass just under the spot [NORDEND 08.10 02:15]; or nothing [LENZSPITZE 27.10 22:00].
  • Over time: hours of sliding; when the obstacle recedes all at once, the spot reopens and the target comes back in one step toward its neighbours, then the sliding resumes [CLARIDEN 18.09 00:00]. Elsewhere an hour [LENZSPITZE 27.10 22:00], three days [WILDSTRUBEL 03.09 02:45], a week with a partial return [WILDSTRUBEL 24.09 02:30], three weeks in steps [NORDEND 08.10 02:15]; the same obstruction returns at the same hour of the night on the same targets [WILDSTRUBEL 08.09 02:30].
  • The cause, and how it was known: seen on the images (the cut spot); the nature of the obstacle is not established [CLARIDEN 18.09 00:00], nor that of the mark [NORDEND 08.10 02:15]; nothing is visible, according to the expert [LENZSPITZE 27.10 22:00, LENZSPITZE 29.10 17:00].
  • What is concluded: the structure does not move, but what is in front of the target is to be checked; the sliding is the artifact, the jump is the return to normal [CLARIDEN 18.09 00:00]; clean the glass [NORDEND 08.10 02:15].
  • Variants: the error_value may return to its level without the curve returning [NORDEND 08.10 02:15]; a vehicle that hides only part of the spot makes the same sliding [BREITHORN 23.10 16:00].
  • Labels: CLARIDEN 18.09 00:00; WILDSTRUBEL 03.09 02:45, WILDSTRUBEL 24.09 02:30, WILDSTRUBEL 02.10 12:45, NORDEND 08.10 02:15, LENZSPITZE 27.10 22:00, WILDSTRUBEL 08.09 02:30, and 2 others: WILDSTRUBEL 1, LENZSPITZE 1; [E-010], D-303-A of CLARIDEN 18.09 00:00 (the return jump around 15:30).

The soiled glass

  • What happens: after rain, drop marks remain on the glass in front of the lens and veil part of the scene, durably.
  • What is seen: the corrections of a few targets next to a mark stay shifted, their error_value a little higher; the corrections of the references change level, sigma0 higher; blurred discs on the glass, visible every night, absent before the rain.
  • Over time: appears at the end of a rain; does not come back until the end of the batch.
  • The cause, and how it was known: seen on the images.
  • What is concluded: it is not a movement; clean the glass. A mark under a single spot is an obstacle (previous card).
  • Labels: NORDEND 25.10 05:45 (one station).

Rain

  • What happens: drops in front of the lens veil the spots during the hours of rain, at night above all [NIEDERHORN 01.10 04:00, NIEDERHORN 02.10 02:00], by day too [JUNGFRAU 24.10 21:00].
  • What is seen:
    • targets: the error_value of most of them rises clearly, without an isolated peak, then comes down progressively [NIEDERHORN 01.10 04:00]; many rejected at times, the others fidget [NIEDERHORN 13.10 08:00]; the targets no longer move as a block and their corrections slide [JUNGFRAU 24.10 21:00];
    • references: they hold, higher than the targets, "probably less affected by the angle effect"; two rise a little at the end of the night, far from the threshold [NIEDERHORN 01.10 04:00, NIEDERHORN 02.10 02:00]; elsewhere, rejected in turn: the camera is often no longer realigned and the measurements kept jump together [NORDEND 24.10 16:30, FLORIDA 24.10 21:45]; on a station with two references, almost all the curves are missing [FLORIDA 01.10 20:45];
    • images: veiled and noisy background, spots that shrink; drops lit by the floodlight (image of 08:00) [NIEDERHORN 02.10 02:00] or by the infrared, as blurred discs [NORDEND 24.10 16:30]; by day, the wet road [JUNGFRAU 24.10 21:00], the leaves covered with drops [FLORIDA 01.10 20:45], the low sky [NORDEND 23.10 08:45].
  • Over time: several hours at night, progressive return; or a day and the night that follows [NORDEND 23.10 08:45]; return all at once at the switch to infrared at nightfall [JUNGFRAU 24.10 21:00]; a target sometimes comes back after the others [FLORIDA 24.10 21:45].
  • The cause, and how it was known: seen: the drops [NIEDERHORN 02.10 02:00], the wet road [JUNGFRAU 24.10 21:00]; otherwise "rain or mist, probable", from fainter spots without an obstacle [NIEDERHORN 25.10 03:00]. Rain, mist and fogging cannot be told apart on the measurements [E-001].
  • What is concluded: the structure does not move, nothing to do; curves of the range not to be trusted [FLORIDA 24.10 21:45]; no valid measurement when the references fall [NORDEND 24.10 16:30]. After the rain, marks may remain (see "The soiled glass").
  • Labels: NIEDERHORN 01.10 04:00, NIEDERHORN 02.10 02:00; JUNGFRAU 24.10 21:00, FLORIDA 24.10 21:45, NORDEND 24.10 16:30, NORDEND 23.10 08:45, FLORIDA 01.10 20:45, NIEDERHORN 29.10 00:00, and 7 others: NIEDERHORN 5, FLORIDA 2; [E-005], the rainy nights of NIEDERHORN 01.10 04:00 and 02.10 02:00.

Visibility drops over the whole scene, without an obstacle

  • What happens: the whole scene is seen less well, with nothing in front of the targets [JUNGFRAU 07.10 08:36].
  • What is seen:
    • targets: the error_value of most of them and of one reference rises clearly; the curves become noisier, with apparent movements that are not real [JUNGFRAU 07.10 08:36]; a target rejected almost all the time disappears from the curves [JUNGFRAU 18.10 10:00]; the corrections of most of them jump together several times [JUNGFRAU 23.10 07:30]; on a recent camera, the flux drops strongly over almost the whole scene, unevenly, the image of the faint targets is ambiguous and the big nearby spots jump between two positions [LENZSPITZE 18.10 08:00];
    • references: they rise without being dropped, but the raw position of one drifts away and distorts the pose [JUNGFRAU 20.10 14:15]; one that passes just above the threshold is dropped and the whole base jumps [NORDEND 04.09 10:59]; two lost, only one remains [NORDEND 14.10 05:00];
    • temperatures: the obstruction sometimes comes with a drop in the temperature of the housing [CLARIDEN 06.10 18:00, LENZSPITZE 21.10 02:50];
    • images: nothing distinct, the spots bigger and more blurred [JUNGFRAU 07.10 08:36], or a little less sharp [NORDEND 04.09 10:59]; a veil, milky scene, halos [FLORIDA 23.10 12:15, JUNGFRAU 18.10 10:00]; all the spots shrinking [NORDEND 14.10 05:00, FLORIDA 18.10 05:00].
  • Over time: more than a day [JUNGFRAU 07.10 08:36]; hours, at night or by day [FLORIDA 18.10 05:00, LENZSPITZE 17.10 10:55]; or a single measurement, all the sensors together [NORDEND 04.09 10:59]. Return all at once [JUNGFRAU 01.10 11:15], or when the mist lifts [NORDEND 14.10 05:00].
  • The cause, and how it was known: not established; perhaps wind [JUNGFRAU 07.10 08:36] (the automatic label of the same range, JUNGFRAU 07.10 08:30, sees a grey scene and a wet road and says "rain or mist": the images show an overcast day, without an obstacle, nothing decisive); a mist seen [NORDEND 10.09 04:30, NORDEND 14.10 05:00, FLORIDA 23.10 12:15, NIEDERHORN 14.10 11:00] or probable [JUNGFRAU 18.10 10:00]; water on the lens or the prisms, possible [LENZSPITZE 17.10 10:55].
  • What is concluded: the visibility is worse; the structure does not move, nothing to do [JUNGFRAU 07.10 08:36]; do not trust the curves of the range [JUNGFRAU 18.10 10:00]; no valid measurement when only one reference remains [NORDEND 14.10 05:00].
  • Variants:
    • thermal effects are added to it and make the case hard to read [JUNGFRAU 07.10 08:36, JUNGFRAU 20.10 14:15];
    • a single measurement: all the spots a little less sharp, references included, positions and corrected curves intact: "atmospheric conditions", very slight, nothing to mask [E-007], an instant of NORDEND not labeled separately (01.09, under the thermal label NORDEND 01.09 06:45); stretched spots, perhaps a motion blur, a reference dropped and the base jumping [NORDEND 28.08 12:30]; a shot too dark, almost black, references dropped, a failed shot, not an obstacle [WILDSTRUBEL 06.10 14:15, WILDSTRUBEL 04.10 14:00];
    • in a tunnel: the same bump returns on the same group of targets, each in its own direction, with the error_value of many targets, without camera movement, images still dark; perhaps the air of the tunnel [CLARIDEN 01.10 06:15, CLARIDEN 06.10 18:00, CLARIDEN 02.10 21:45].
  • Labels: JUNGFRAU 07.10 08:36, NORDEND 04.09 10:59; NORDEND 14.10 05:00, FLORIDA 23.10 12:15, LENZSPITZE 18.10 08:00, CLARIDEN 06.10 18:00, WILDSTRUBEL 06.10 14:15, JUNGFRAU 20.10 14:15, and 35 others: LENZSPITZE 8, NORDEND 8, CLARIDEN 7, JUNGFRAU 4, WILDSTRUBEL 4, FLORIDA 2, NIEDERHORN 1, SNCF01 1; [E-007].

Part of the scene goes dark at night

  • What happens: at night, only part of the scene loses its light, without an obstacle: the distant spots [LENZSPITZE 16.10 22:25], or a few neighbouring targets at the edge of the group [NORDEND 30.09 20:45].
  • What is seen: the targets of the affected part rejected for a large part of the night [LENZSPITZE 16.10 22:25, NORDEND 16.10 19:00], the corr_x of the distant ones drifting before the first rejections [LENZSPITZE 22.10 19:40]; the others move a little with the changes of base [LENZSPITZE 16.10 22:25], or not at all [NORDEND 16.10 19:00]. The distant references rejected or dropped in turn; one, just under its threshold, measured at times on a false position, its corrections absurd [LENZSPITZE 16.10 22:25]. On a recent camera, the flux of the distant ones falls to almost nothing from the start of the evening, well before the curves, that of the near ones stays whole [LENZSPITZE 16.10 22:25]. Images: the distant spots pale, the near ones stay sharp [LENZSPITZE 22.10 19:40]; at the edge of the group, a few spots shrink then reopen [NORDEND 30.09 20:45].
  • Over time: a night, return in the early morning [LENZSPITZE 16.10 22:25]; night after night, not all of them [NORDEND 16.10 19:00]; sometimes by day, under an overcast sky, a few measurements [NORDEND 21.10 10:15].
  • The cause, and how it was known: not established; mist or humidity at night, probable [LENZSPITZE 16.10 22:25] or possible [NORDEND 16.10 19:00].
  • What is concluded: it is not a movement; ignore the curves of the range, those of the badly measured reference above all [LENZSPITZE 16.10 22:25].
  • Variants: the worst night, the near ones also lose their light and the compensation no longer keeps any reference (see "Two references lost") [LENZSPITZE 26.10 00:00].
  • Labels: LENZSPITZE 16.10 22:25, LENZSPITZE 26.10 00:00, LENZSPITZE 22.10 19:40, NORDEND 30.09 20:45, NORDEND 16.10 19:00, NORDEND 21.10 10:15, and 11 others: NORDEND 8, LENZSPITZE 3.

A reference loses its brightness, without an obstacle

  • What happens: the spot of one or two references weakens and shrinks with nothing in front, most often at night, then comes back on its own; the other sensors stay well seen [E-001, JUNGFRAU 13.10 23:45].
  • What is seen:
    • references: the error_value of the affected ones rises, all at once or little by little, the other staying at its level; under the threshold, none is dropped, but the reprojection increases and the computed pose changes while the raw positions of the targets do not move [JUNGFRAU 13.10 23:45, JUNGFRAU 02.10 01:45]; above it, it is dropped (see D) [JUNGFRAU 26.10 01:56];
    • targets: error_value at its level, raw positions without a step; their corrections move only through the compensation: small common shifts of the corr_x of a few targets, at the entry and at the exit of the reference, which do not always find their level again before daylight [E-001, JUNGFRAU 13.10 01:00, JUNGFRAU 22.10 00:15];
    • images: the spot reduced to a point when that of a witness target stays whole, nothing in front [E-001, JUNGFRAU 22.10 00:15, JUNGFRAU 26.10 00:00]; sometimes the image a little veiled, a halo [JUNGFRAU 13.10 23:45].
  • Over time: a few hours at night, return on its own before daylight [E-001]; several nights in a row [JUNGFRAU 26.10 00:00, JUNGFRAU 26.10 23:00, JUNGFRAU 28.10 01:30]; or a slow rise all night and the loss morning after morning [FLORIDA 28.10 07:00].
  • The cause, and how it was known: fogging or water, mist in the air or fogging on the reflectors: deduced, not seen, nothing allows deciding [E-001]; elsewhere not established [FLORIDA 23.10 03:45, JUNGFRAU 13.10 23:45].
  • What is concluded: the mist veils the spot without moving its centre: no false movement; the small shifts of the compensation are negligible (review of the event); healthy system, nothing to do [E-001]; they are not a movement [JUNGFRAU 13.10 01:00]. Dropped or two lost: the cards of D; if it repeats, check the prisms [JUNGFRAU 27.10 00:56].
  • Variants: by day, a tiny reference passes the threshold for a few measurements; on two references, the camera is no longer realigned [FLORIDA 02.10 09:45]; before its loss, the reference slides in the image against the targets and distorts the pose, the temperature of the housing staying flat [FLORIDA 27.10 01:45]; the base that hesitates between two levels from one measurement to the next, the spot whole, cause not established [JUNGFRAU 21.10 12:45]; in a tunnel, a fainter reference, rejected at times [CLARIDEN 31.10 22:15].
  • Labels: JUNGFRAU 13.10 23:45, JUNGFRAU 26.10 00:00, JUNGFRAU 22.10 00:15, FLORIDA 28.10 07:00, FLORIDA 27.10 01:45, CLARIDEN 31.10 22:15, and 26 others: JUNGFRAU 12, FLORIDA 11, NORDEND 3; [E-001], the night of JUNGFRAU 13.10 23:45; PS_DES_COUDRES 13.10 23:45, awaiting review, is the same night of the same station (section 4).

Atmospheric turbulence

  • What happens: layers of air at different temperatures disturb the image; a physical effect that cannot be worked around [WILDSTRUBEL 27.09 04:15]; air heated above a road [E-003], above ballast and rails [LENZSPITZE 16.10 11:00], above tracks [SNCF01 04.10 08:15] makes the spots tremble.
  • What is seen: by day, disturbances of the error_value, globally well compensated [WILDSTRUBEL 27.09 04:15]; the corrected curves of all the targets fidget much more than at night, one-measurement jumps that come straight back, without a step; a few targets rejected at times [SNCF01 04.10 08:15]; a few single-step jumps, mostly on the most distant sensors [E-003]. At the peak of sunrise, the compensation drops all the references for a few measurements [WILDSTRUBEL 04.09 03:45]. On a recent camera, the flux does not drop, the image is less sharp: it trembles, it does not darken [LENZSPITZE 16.10 11:00]. Images: full sun, sharp shadows, nothing in front [LENZSPITZE 16.10 11:00]; turbulence cannot be seen on a still image [E-003].
  • Over time: by day, from sunrise to sunset, according to the images [WILDSTRUBEL 27.09 04:15]; every sunny day, calm at night [SNCF01 04.10 08:15], weaker some days [SNCF01 09.10 09:15].
  • The cause, and how it was known: not stated; the expert gives it as a trait of the station [WILDSTRUBEL 27.09 04:15]; elsewhere deduced from a fast noise over the whole scene, by day only, in sunshine [SNCF01 04.10 08:15, E-003].
  • What is concluded: the structure does not move, nothing to do; do not trust the detail of the curves [LENZSPITZE 16.10 11:00].
  • Variants: the low sun at sunrise is added to it, roadway covered with reflections [WILDSTRUBEL 07.09 03:30]. On the most affected target, the corr_x hollows out by day: for the expert, the cycle of the event, stronger on this target, not a movement [E-003, PB1_2 of WILDSTRUBEL]; automatic labels read the targets at the bottom of the image (PB1_2, PB2_2) as badly compensated in the morning [WILDSTRUBEL 24.09 04:00]. Daytime agitation, calm camera: perhaps turbulence, not established [NORDEND 13.09 11:30].
  • Labels: WILDSTRUBEL 27.09 04:15; WILDSTRUBEL 01.09 03:45, WILDSTRUBEL 04.09 03:45, LENZSPITZE 16.10 11:00, LENZSPITZE 22.10 07:40, SNCF01 04.10 08:15, SNCF01 16.10 08:15, and 48 others: WILDSTRUBEL 25, SNCF01 19, LENZSPITZE 3, NORDEND 1; [E-003], the days of WILDSTRUBEL 01.09 03:45 to 04.09 03:45.

The sun in the field or behind a spot

  • What happens: the light of the sun drowns a spot: the sun enters the field [NIEDERHORN 31.10 11:00], or it lights the background of a target or a reference, as bright as it is [NORDEND 12.08 13:30, JUNGFRAU 12.10 15:45].
  • What is seen: a single target rejected, its error_value rising from the end of the morning then dropping all at once when the wall returns to shade [NORDEND 12.08 13:30]; a target at the edge of a light border whose corr_x drifts away all day [WILDSTRUBEL 02.09 04:15]. The sun in the field: the error_value of the references rises without rejection, the compensation a little disturbed, the corrections of many targets jump or fidget [NIEDERHORN 22.10 12:00]; a reference dropped for one measurement [NIEDERHORN 31.10 11:00]. The background of a reference in full sun: small common jump of several targets, under the threshold [JUNGFRAU 12.10 15:45]; or reference dropped for hours, its neighbours in the image on a step (see "A reference dropped") [WILDSTRUBEL 08.09 04:00, WILDSTRUBEL 08.10 06:00]. Images: the sun in the field, halo and reflections [NIEDERHORN 31.10 11:00]; the rock or the pavement in full sun around the spot [NORDEND 12.08 13:30]; sharp in the shade and at night [WILDSTRUBEL 02.09 04:15].
  • Over time: at the same hours every sunny day, longer in August than in October [NORDEND 05.10 14:45]; intermittently on a cloudy day [NORDEND 15.08 13:00]; a few measurements, sometimes a single one, at sunrise [NIEDERHORN 22.10 12:00].
  • The cause, and how it was known: seen on the image of the day [NORDEND 12.08 13:30, JUNGFRAU 12.10 15:45]; deduced on the other days, same hour and same signature [NORDEND 14.09 14:15]; a dazzling sky behind the clouds, probable [NIEDERHORN 26.10 13:00].
  • What is concluded: the structure does not move, nothing to do; the curve of the target is missing or not to be trusted at those hours [NORDEND 12.08 13:30]; the steps of a dropped reference are not a movement [WILDSTRUBEL 08.10 06:00].
  • Variants: grazing sun, the top of the image lit, targets then two references rejected and all the targets jumping [NORDEND 30.10 08:45]; the moving shadow of a tree makes a target left under the threshold latch onto something else [WILDSTRUBEL 24.09 05:30]; on a road in full sun, parasitic spots where the detection jumps for one measurement [E-003]; the rising sun on the wall, at the moment when the housing heats up all at once, makes the pose of the camera jump [NORDEND 11.08 05:45].
  • Labels: NIEDERHORN 31.10 11:00, NORDEND 12.08 13:30, JUNGFRAU 12.10 15:45, WILDSTRUBEL 02.09 04:15, WILDSTRUBEL 08.09 04:00, NORDEND 30.10 08:45, and 84 others: NORDEND 42, NIEDERHORN 18, WILDSTRUBEL 18, JUNGFRAU 6; [E-003].

A faint spot lost when the light is missing

  • What happens: a target whose spot is already faint goes above its threshold when the light is missing: transition from night to day [FLORIDA 01.10 07:15], overcast sky [NORDEND 16.08 09:15], noisy night image [NORDEND 08.09 01:45].
  • What is seen: a single target, or the faintest ones, rejected for one to a few measurements, the others motionless [FLORIDA 01.10 07:15, NORDEND 16.08 09:15]; a few neighbours make a small jump [NORDEND 03.09 13:15]. Images: the spot shrinks as the day rises when that of a neighbour stays wide [FLORIDA 01.10 07:15]; dark scene, spot that can hardly be seen [NORDEND 16.08 09:15].
  • Over time: one to a few measurements; sometimes an overcast morning [NORDEND 17.08 02:30], a night, several nights in a row [SNCF01 17.10 20:00, SNCF01 18.10 17:00].
  • The cause, and how it was known: seen on the images [FLORIDA 01.10 07:15]; elsewhere not established [NORDEND 16.08 09:15]; a prism of little brightness, with insufficient contrast by day (case: expert-annotated-cases.md, GN-1.472).
  • What is concluded: it is not a movement; points to ignore [NORDEND 16.08 09:15]. Boundary with C: a target that has its own rejections almost every day, also in overcast or rainy weather, is a problem target; a target right at the threshold that crosses it only now and then, its rejections coming with the light, is here [SNCF01 19.10 13:30, SNCF01 30.10 21:30, FLORIDA 02.10 10:30].
  • Labels: FLORIDA 01.10 07:15, NORDEND 16.08 09:15, NORDEND 08.09 01:45, SNCF01 19.10 13:30, SNCF01 17.10 20:00, NORDEND 17.08 02:30, and 22 others: NORDEND 16, SNCF01 4, FLORIDA 2.

C. The target has changed

A target rejected at all its measurements

  • What happens: a target is never accepted, whatever the weather; its corrected curve does not exist.
  • What is seen: rejected at every measurement, at most a few isolated points; its error_value almost at 1 [CLARIDEN 01.10 00:00 (P-300-B)], or at the level of its neighbours and following their variations, but above a threshold of its own, lower than that of the others [JUNGFRAU 30.09 00:00 (PM_G)]. Images, depending on the target: no spot [NIEDERHORN 01.10 00:00 (e03)]; a sharp spot next to the position where the system looks for it [CLARIDEN 01.10 00:00 (P-300-B)]; between two very close neighbours [NIEDERHORN 01.10 00:00 (e07)]; very large and dazzled at night, stuck to a neighbour [NIEDERHORN 01.10 00:00 (e11)]; doubled [SNCF01 04.10 00:00 (BG.1)]; sharp and round like its neighbours [JUNGFRAU 30.09 00:00 (PM_G)].
  • Over time: permanent, sometimes since the start of the data [JUNGFRAU 30.09 00:00 (PM_G)]; or since a setting (see "The configuration of a sensor has changed") [CLARIDEN 01.10 00:00 (P-300-B)].
  • The cause, and how it was known: the look of the spot seen on the thumbnails; the cause not established (configuration, threshold, prism); sharp spot: rather a badly set threshold [JUNGFRAU 30.09 00:00 (PM_G)].
  • What is concluded: it is not the weather; ignore its curve; act on the target or its configuration.
  • Labels: CLARIDEN 01.10 00:00 (P-300-B), JUNGFRAU 30.09 00:00 (PM_G), NIEDERHORN 01.10 00:00 (e03), NIEDERHORN 01.10 00:00 (e07), NIEDERHORN 01.10 00:00 (e11), SNCF01 04.10 00:00 (BG.1); and 5 others: NIEDERHORN 4, SNCF01 1. PM_G is also rejected throughout [E-001].

A target right at the threshold, the whole batch

  • What happens: a target lives permanently at the limit of detection: its error_value keeps crossing the threshold.
  • What is seen: a large part of its measurements rejected, by day as by night, in all weathers; its curve full of holes, its points kept noisier [SNCF01 04.10 00:00 (F4.3)] or jumping [FLORIDA 01.10 00:00 (Sellette 6)]; its last accepted points may drift away strongly without anything else moving [SNCF01 04.10 00:00 (CV.7)]. Images: a small spot very close to a much brighter reflector, onto which its measurement may latch [FLORIDA 01.10 00:00 (Sellette 6), SNCF01 04.10 00:00 (CV.6)]; faint and doubled [SNCF01 04.10 00:00 (F4.4)]; at the edge of a light structure [SNCF01 04.10 00:00 (CV.7)].
  • Over time: the whole batch; the error_value may rise slowly, rejections in runs of several hours, at night above all [SNCF01 04.10 00:00 (CV.6)].
  • The cause, and how it was known: the look of the spot seen on the images; not established (orientation, cleanliness of the prism).
  • What is concluded: do not trust its curve; check the orientation or the cleanliness of the prism. With very few own rejections, it is not this phenomenon (see the previous card of B).
  • Labels: FLORIDA 01.10 00:00 (Sellette 6), SNCF01 04.10 00:00 (CV.7), SNCF01 04.10 00:00 (F4.4), SNCF01 04.10 00:00 (F4.3), SNCF01 04.10 00:00 (CV.6).

A target rejected every night or every day

  • What happens: a target goes above its threshold under the ordinary regime of the station (floodlight at night, daylight), at the same hours, whatever the weather.
  • What is seen: rejected almost every night [NIEDERHORN 01.10 00:00 (PWD_RMP_B15_017_C09), NIEDERHORN 15.10 01:00], or every day [NIEDERHORN 01.10 00:00 (e01), WILDSTRUBEL 01.09 00:00]; measured the rest of the time, its points just under the threshold more agitated [NIEDERHORN 01.10 00:00 (e01)]. Images: at night, a saturated spot spread in a cross under the floodlight, small and sharp by day [NIEDERHORN 01.10 00:00 (PWD_RMP_B15_017_C09)]; by day, at the edge of the image, almost melted into the background [NIEDERHORN 01.10 00:00 (e01)]; a reference invisible by day, sharp at night under the infrared [WILDSTRUBEL 01.09 00:00]; a sharp spot on a lit ballast background [NIEDERHORN 01.10 00:00 (PWD_IMP_B08_024_C09)].
  • Over time: the whole batch or from a date [NIEDERHORN 17.10 03:00]; the rejections may become rarer [NIEDERHORN 01.10 00:00 (PWD_IMP_B08_024_C09)] or stop, the curve staying more scattered than that of the neighbours [NIEDERHORN 01.10 00:00 (PWD_RMP_B15_017_C09)].
  • The cause, and how it was known: deduced from the rejections at the same hours whatever the weather; for a reference, also the days when the housing heats the least: it is the target, not the sun [WILDSTRUBEL 07.09 00:00, WILDSTRUBEL 01.10 00:00]. The reason is not established.
  • What is concluded: it is not a movement; ignore its curve at the affected hours; review threshold, orientation, lighting or target. A reference cleaned or replaced gives one more reference to the daytime compensation; its entries and exits do not make the targets jump [WILDSTRUBEL 01.09 00:00].
  • Labels: NIEDERHORN 01.10 00:00 (PWD_RMP_B15_017_C09), NIEDERHORN 01.10 00:00 (e01), NIEDERHORN 15.10 01:00, WILDSTRUBEL 01.09 00:00, WILDSTRUBEL 01.10 00:00, NIEDERHORN 17.10 03:00; and 2 others: NIEDERHORN 1, WILDSTRUBEL 1.

A target that degrades, and sometimes recovers

  • What happens: a target that used to be seen well is seen worse and worse from a date on, and does not come back, or only after weeks.
  • What is seen: the error_value changes level and stays there [NIEDERHORN 30.10 01:00 (PWD_RMP_B16_026_C09)], or rises over the weeks up to the threshold then above [NORDEND 16.10 19:00 (CTR01), NORDEND 17.08 05:15]; the rejections first by day, then by day as by night [FLORIDA 06.10 10:15]; between them, noisier curves that change level on their own [NORDEND 17.08 05:15]. Images: a spot that weakens then becomes invisible by day [FLORIDA 06.10 10:15]; much fainter than its neighbours [NORDEND 17.08 05:15]; blurred marks on the glass [NORDEND 16.10 19:00 (CTR01)]; the same spot as before [NIEDERHORN 30.10 01:00 (PWD_RMP_B16_026_C09)].
  • Over time: the change starts from an episode: a rain after which the spot does not come back [FLORIDA 06.10 10:15], a jump of the neighbours at the same measurement [NIEDERHORN 30.10 01:00 (PWD_RMP_B16_026_C09)], flying insects in front of the lens at the instant when three neighbouring targets change error_value in opposite directions, one rejected afterwards for days [E-013]; then it lasts. Or it lasts weeks until a rain after which the error_value falls back and the rejections stop: the target probably cleaned itself [NORDEND 01.10 00:00 (PR04)].
  • The cause, and how it was known: not established (prism dulled, soiled or veiled, glass, orientation, web, or a change around the target).
  • What is concluded: ignore its curve; check the target. A kept value is not a correct value: kept again under the threshold, the target may stay far from its previous level [FLORIDA 06.10 10:15]; recovered, it stays lower than at the start of the month [NORDEND 01.10 00:00 (PR04)].
  • Variants: a false point: the position leaves a weakened spot, under the threshold, and the point stays in the curve [CLARIDEN 27.10 12:00].
  • Labels: FLORIDA 06.10 10:15, NORDEND 17.08 05:15, NORDEND 16.10 19:00 (CTR01), NIEDERHORN 30.10 01:00 (PWD_RMP_B16_026_C09), CLARIDEN 27.10 12:00, NORDEND 01.10 00:00 (PR04); and 1 other: NIEDERHORN 1. [E-008] is PR04 of NORDEND on 01 and 02.09, at the start of NORDEND 17.08 05:15: a dull target, sharp at night, almost melted into the background by day when its neighbours saturate, its corrections more agitated without a step; validated "state of the target to be checked", its rejections multiplying afterwards. [E-013] (learned, not revalidated) is PWD_IMP_B08_024_C09 of NIEDERHORN on 26.09 in the morning, before the labels: the start of the rejections of NIEDERHORN 01.10 00:00 (PWD_IMP_B08_024_C09).

The configuration of a sensor has changed

  • What happens: a sensor is reconfigured, or a setting changes (sensor moved, starting position changed); its curve changes without anything moving in the image.
  • What is seen: a starting position changed: raw positions flat, relative displacement that jumps; two sensors of the same prism jump in opposite directions and stay there [CLARIDEN 29.10 23:45]. Two sensors placed on the same point of a group of spots, same positions and same curves, separate at one measurement, each on a part of the group, and their curve joins that of the neighbours, camera motionless [MICHIGAN 03.10 00:00]. A setting changed one day: a target stays next to its spot and is rejected since, the error_value of other targets changing level at the same hour [CLARIDEN 01.10 00:00 (P-300-B)].
  • Over time: all at once; the new state stays.
  • The cause, and how it was known: deduced from the raw positions and the thumbnails; "the sensors were reconfigured", not established [MICHIGAN 03.10 00:00].
  • What is concluded: not a movement; do not compare before and after; check with the operator. The jump may be the return to normal [MICHIGAN 03.10 00:00]. Until then, two sensors on the same point count each episode twice.
  • Labels: MICHIGAN 03.10 00:00, CLARIDEN 29.10 23:45, CLARIDEN 01.10 00:00 (P-300-B).

D. The references

A reference dropped: the base changes

  • What happens: a reference is seen badly, its error_value exceeds the threshold, the system drops it. With two references, there is always a transformation that fits them without error, but very different from the one with three: the base changes.
  • What is seen:
    • the reference: its error_value jumps almost to 1 (hidden) [JUNGFRAU 06.10 05:45], rises little by little [JUNGFRAU 26.10 01:56], or passes just above the threshold [NORDEND 04.09 10:59]; its raw position jumps when it latches onto a reflection [WILDSTRUBEL 23.09 09:15]; on a recent camera, its light being cut says it is hidden even when the image of the minute shows nothing [LENZSPITZE 17.10 05:00];
    • the reprojection error falls to zero; the corrections of the remaining references too [NORDEND 04.09 10:59] (the data confirm it, at this measurement as at NORDEND 28.08 12:30 and NORDEND 18.08 17:45, where the automatic labels say they "also jump": they jump to zero);
    • the targets: all jump together, raw positions without a step [NORDEND 28.07 20:45]; one measurement for a pass-by, a step of several hours when the reference stays hidden [NORDEND 18.08 22:15], above all for its neighbours in the image [WILDSTRUBEL 08.09 04:00];
    • images: the spot reduced to a point when a witness target stays whole [JUNGFRAU 28.10 01:30]; a pass-by, seen or in transparency [WILDSTRUBEL 25.09 12:00]; an animal [NIEDERHORN 09.10 15:00]; a shot too dark [WILDSTRUBEL 06.10 14:15]; a faint spot on a light background [WILDSTRUBEL 08.09 04:00].
  • Over time: one measurement or hours; everything comes back when it reappears, sometimes with a small rebound [NORDEND 18.08 22:15].
  • The cause, and how it was known: that of the loss: a truck seen on the image of 06:00 [JUNGFRAU 06.10 05:45]; a web [NORDEND 28.07 20:45]; a visibility that drops, cause not stated [JUNGFRAU 26.10 01:56] or unknown, nothing hiding the reference [NORDEND 04.09 10:59]; a pass-by deduced from its signature [WILDSTRUBEL 28.09 05:15]; the sun [NIEDERHORN 31.10 11:00, WILDSTRUBEL 08.10 06:00]. The mechanism can be read on the reprojection at zero.
  • What is concluded: the jumps are not a movement; it is a very bad sign for the measurement: the system runs on two references when it needs three [JUNGFRAU 26.10 01:56].
  • Variants: depending on the reference dropped, the jump is clear or almost nil [NORDEND 28.07 20:45]; on a station that has more than three, all the targets jump [WILDSTRUBEL 03.10 07:30], a few [BREITHORN 07.10 08:00], or nothing [BREITHORN 08.10 20:00]. A reference merely disturbed, without being dropped, makes a good part of the targets move slightly [NORDEND 04.09 16:09]: the corrections of all of them flip from one measurement to the next [JUNGFRAU 21.10 12:45]; an object in front of it hides it without it being dropped, and all the targets jump [LENZSPITZE 30.10 13:50]. A rejected reference may stay counted in the compensation [LENZSPITZE 23.10 16:55].
  • Labels: JUNGFRAU 06.10 05:45, JUNGFRAU 26.10 01:56, NORDEND 04.09 10:59, NORDEND 28.07 20:45, NORDEND 04.09 16:09; JUNGFRAU 28.10 01:30, NORDEND 18.08 22:15, WILDSTRUBEL 08.09 04:00, NIEDERHORN 09.10 15:00, LENZSPITZE 30.10 13:50, BREITHORN 07.10 08:00, and some 80 others, mostly WILDSTRUBEL, then LENZSPITZE, NORDEND, JUNGFRAU, BREITHORN; [E-002].

Two references lost: the camera is blind

  • What happens: two of the three references become almost invisible, and the system drops them one after the other; sometimes the compensation no longer keeps any.
  • What is seen: their error_value rises strongly. The reprojection error rises first, the camera struggling to realign, falls to zero at the loss of the first, and is no longer computed at all at the loss of the second (missing values) [JUNGFRAU 14.10 23:02, JUNGFRAU 27.10 00:56]. Targets: jumps in corr_x and corr_y at each switch, then a movement, displacements less and less reliable; when a reference returns, the reprojection passes through zero again, then everything comes back [JUNGFRAU 14.10 23:02]. With no reference kept: sigma0 is missing, the corrections of the references still well seen stay frozen, the curves that remain jump and drift together [LENZSPITZE 26.10 00:00]; during a pass-by, the targets left under the threshold jump [SNCF01 05.10 15:45]. Images: at night, without an obstacle, the spots of the two references reduced to points, a witness target whole [JUNGFRAU 26.10 23:00]; the distant spots almost all gone [LENZSPITZE 26.10 00:00].
  • Over time: a few hours, one reference after the other [JUNGFRAU 14.10 23:02]; the length of a pass-by [SNCF01 05.10 15:45]; a night [LENZSPITZE 26.10 00:00].
  • The cause, and how it was known: the poor visibility of the two references, read on their error_value and on the reprojection; its own cause is not stated [JUNGFRAU 14.10 23:02]; probable elsewhere: mist or humidity [LENZSPITZE 26.10 00:00], mist seen [NORDEND 14.10 05:00], rain [NORDEND 24.10 16:30]; all lost at once: a train [SNCF01 05.10 15:45], parked cars [WILDSTRUBEL 26.09 03:45 (Obstruction)]; at the peak of sunrise, a few measurements [WILDSTRUBEL 04.09 03:45].
  • What is concluded: with a single reference, the camera is blind: every measurement of the range is invalid; with none, the points have no meaning [WILDSTRUBEL 15.10 07:45]. If the problem repeats, check the state of the reference prisms [JUNGFRAU 27.10 00:56].
  • Variants: under the jumps, the underlying trend stays the thermal effect on the bridge [JUNGFRAU 14.10 23:02].
  • Labels: JUNGFRAU 27.10 00:56, JUNGFRAU 14.10 23:02; JUNGFRAU 26.10 23:00, LENZSPITZE 26.10 00:00, SNCF01 05.10 15:45, WILDSTRUBEL 15.10 07:45, NORDEND 14.10 05:00, NORDEND 24.10 16:30, and some thirty others, mostly SNCF01 and WILDSTRUBEL.

A compensation on two references

  • What happens: the compensation runs on only two references, the station having only two or having lost one for good; fitted without error, it holds the camera badly and passes on to all the targets the slightest difference between them.
  • What is seen: all the targets move together at the slightest difference between the two references, without their raw positions moving, the pose flipping back and forth [MICHIGAN 26.10 20:00, MICHIGAN 17.10 12:00]; a movement of the camera badly compensated, the scene deforming [NEBRASKA 16.10 20:00 (drift)]. One of the two rejected: the camera is no longer realigned, several targets jump a little and come back at the next realignment [FLORIDA 02.10 09:45]; badly seen without being rejected, the pose is distorted [FLORIDA 23.10 22:00]. At the loss for good, all the targets jump in one step and stay shifted; the reprojection and the corrections of the two remaining cancel; the lost spot has disappeared from its place, without an obstacle [MICHIGAN 09.10 08:00, NEBRASKA 09.10 08:00]. On a station that has only two references, the reprojection is permanently at zero, inlier and sigma0 empty [FLORIDA 01.10 20:45, FLORIDA 22.10 11:15].
  • Over time: permanently, or since the loss, without return [MICHIGAN 09.10 08:00]; a reference that weakens every night and is lost every morning [FLORIDA 23.10 03:45, FLORIDA 28.10 07:00]; an agitation of a few measurements that returns during the day [MICHIGAN 17.10 12:00].
  • The cause, and how it was known: deduced from the pose, the raw positions and the corrections of the references at zero; the loss not established (prism masked, turned, removed). Two stations of the same building, according to the expert, each lose a reference at the same measurement: a common cause is to be looked for [MICHIGAN 09.10 08:00, NEBRASKA 09.10 08:00].
  • What is concluded: it is not a movement; fragile base, very bad sign, unreliable curves; restore a third reference [NEBRASKA 16.10 20:00 (drift)].
  • Variants: a brief thermal movement of the camera, with the cold, badly compensated [NEBRASKA 26.10 00:00]; the twin of a reference follows its reference (see G) [FLORIDA 09.10 09:45].
  • Labels: MICHIGAN 09.10 08:00, NEBRASKA 09.10 08:00, FLORIDA 02.10 09:45, MICHIGAN 26.10 20:00, MICHIGAN 17.10 12:00, FLORIDA 23.10 22:00; and 16 others: FLORIDA 11, NEBRASKA 4, MICHIGAN 1.

The base changes without a reference lost

  • What happens: all the references stay in the compensation, but the distance between them changes a little; all the targets inherit it together, without moving in the image.
  • What is seen: when a reference resumes, all the targets shifted from their previous level; against the overall movement of the targets, a few references have drifted slightly apart [WILDSTRUBEL 01.10 13:00]; the three references and a few targets jump together for one measurement, without a reference dropped or a cause seen [NORDEND 13.09 04:00]; a reference dropped for one measurement, nothing hiding it [NIEDERHORN 28.10 18:00].
  • Over time: a shift taken in a morning of thermal movement and reference switches, which fades in a few days [WILDSTRUBEL 03.10 15:15].
  • The cause, and how it was known: not established: a remainder of the movement of the camera, or a change in the look of the references [WILDSTRUBEL 01.10 13:00].
  • What is concluded: not an established movement; check with the following days.
  • Labels: WILDSTRUBEL 01.10 13:00, WILDSTRUBEL 03.10 15:15, NIEDERHORN 28.10 18:00, NORDEND 13.09 04:00.

E. The camera and its mount

The camera moves through a thermal effect

  • What happens: the lens or the mount of the camera expands and contracts with the temperature; the pose follows, and the compensation by the references erases more or less of it [NORDEND 25.09 16:02, WILDSTRUBEL 27.09 04:15].
  • What is seen:
    • movement of the camera: a slow change, then a fast change, every day. The expert reads on NORDEND the pan and the tilt that "drift from nightfall, then come back all at once in the morning" [NORDEND 25.09 16:02]; the automatic labels name the same shape the other way round: a start all at once at the warming of the morning, a slow return in the evening and at night [NORDEND 25.09 06:00, SNCF01 05.10 06:30]. The data of that night show both: a pan that changes slowly from the afternoon to the middle of the night, flat afterwards, and a fast change early in the morning. The axes each have their own hour: the scale follows the level of the housing temperature, the tilt its warming [WILDSTRUBEL 01.09 03:30]; the tilt and the rotation come back all at once, the pan later, the scale during the night [NIEDERHORN 13.10 13:00];
    • raw positions: all the sensors, references included, move together, as a block [WILDSTRUBEL 01.09 03:30];
    • targets: the same cycle in the corrections, mostly corr_x, when the camera compensates its own movements badly [NORDEND 25.09 16:02]; depending on their place in the image, some hollow out, others rise [NIEDERHORN 13.10 13:00]; nothing when it compensates them well [WILDSTRUBEL 27.09 04:15]; on sunny days, the references drift slightly apart too [NORDEND 02.10 06:00];
    • error_value: ordinary [FLORIDA 07.10 08:45]; housing hot, the pose and the curves fidget from one measurement to the next and the error_value of the targets drops a little [NIEDERHORN 13.10 13:00];
    • temperatures: the pose goes with the housing temperature; without temperature, the cycle is read on the pose and the rhythm of the day [SNCF01 05.10 06:30].
  • Over time: every day, at the rhythm of day and night [NORDEND 25.09 16:02], from sunrise to sunset according to the images [WILDSTRUBEL 27.09 04:15]; without rest at night [NORDEND 02.10 06:00, SNCF01 05.10 06:30], or at rest from the afternoon [WILDSTRUBEL 01.09 03:30]. The cycle does not always close: the pan lower in the morning [NORDEND 22.10 06:30], the camera left at its daytime position [SNCF01 19.10 07:15], two days in one cycle [SNCF01 12.10 08:00]. An outage or an obstruction of the view interrupts it [NORDEND 13.10 07:45].
  • The cause, and how it was known: deduced: the camera moves, and the structure cannot move through a thermal effect (rock face too massive, road); the lens and the mount remain [NORDEND 25.09 16:02, WILDSTRUBEL 27.09 04:15]. Elsewhere, from the pose against the housing temperature [NIEDERHORN 13.10 13:00]; a masonry wall is not supposed to move like that [SNCF01 05.10 06:30]. The sun heats the mount on one side then the other, and the camera tilts a little, compensated: probable, not seen [E-003].
  • What is concluded: the curves deform while the structure does not move: nothing to do [NORDEND 25.09 16:02]. Compensated, it is also labeled, and the label says the compensation absorbs it [FLORIDA 07.10 08:45, NORDEND 21.10 07:15]. What follows a start without return is not explained: do not read it as a movement [SNCF01 19.10 07:15].
  • Variants:
    • on a structure that also moves, the camera fixed on it moves at a similar rhythm [JUNGFRAU 11.10 18:05] or faster than it, its mount heating faster [JUNGFRAU 13.10 07:14]; this difference in rhythm is perhaps the cause of the poor compensation;
    • carried by the night: an overcast day after a warm night, everything comes with the cooling of the evening [NORDEND 01.10 10:00]; a colder night takes the pan beyond its level of the day before [NORDEND 17.10 08:45];
    • overcast day: weak cycle, flat curves [NORDEND 21.10 07:15, LENZSPITZE 30.10 07:20], or axes reversed [NIEDERHORN 12.10 15:00];
    • plateau: a target holds its offset as long as the housing stays hot [LENZSPITZE 19.10 07:00]; a clear cooling takes the targets beyond their night level [LENZSPITZE 31.10 07:40, SNCF01 08.10 08:15];
    • a day / night cycle of the pose, badly compensated, is not always the camera: two references out of four masked by vegetation leave an insufficient base (case: expert-annotated-cases.md, GN-1.472).
  • Labels: NORDEND 25.09 16:02, WILDSTRUBEL 27.09 04:15, JUNGFRAU 11.10 18:05, JUNGFRAU 13.10 07:14; NORDEND 02.10 06:00, WILDSTRUBEL 01.09 03:30, NIEDERHORN 13.10 13:00, SNCF01 05.10 06:30, FLORIDA 07.10 08:45, LENZSPITZE 19.10 07:00, and 192 others: NORDEND 80, NIEDERHORN 28, WILDSTRUBEL 26, FLORIDA 24, SNCF01 22, LENZSPITZE 12; [E-003].

The camera drifts away when the temperature of its housing drops, without a daily cycle

  • What happens: sheltered, in a building or a tunnel, the camera has no daily cycle; it drifts away in jerks when the temperature of its housing drops and comes back when it warms up. It is the temperature of the camera housing that counts, not that of the air around: even sheltered, these drifts are thermal [expert, 2026-11-04, on BREITHORN 19.10 00:00, BREITHORN 23.10 12:00, BREITHORN 28.10 00:00, BREITHORN 28.10 16:00, BREITHORN 29.10 16:00].
  • What is seen: the rotation and the scale drift away, the tilt makes a step [BREITHORN 19.10 00:00]; the camera nods forward, the scale drops [MICHIGAN 28.10 16:00]; in the tunnel, the scale alone [CLARIDEN 06.10 18:40]. Raw positions of all the targets and of the references sliding together [MICHIGAN 28.10 16:00]. Targets: several, and references, drift apart in opposite directions then come back with the camera [BREITHORN 19.10 00:00]; on two references, all together, the drift amplified [MICHIGAN 28.10 16:00]; a peak of the corr_x at the coldest measurement [NEBRASKA 28.10 20:00]; in the tunnel, nothing, the compensation absorbs it [CLARIDEN 06.10 18:40]. error_value: little changed; that of a few sensors a little higher, without rejection, cause not established [NEBRASKA 28.10 20:00]. Nothing obstructs the view.
  • Over time: the cold nights and mornings, not every day; return around noon or in the afternoon [BREITHORN 19.10 00:00]; the same mornings on the cameras of the same building, according to the expert [BREITHORN 26.10 04:00, NEBRASKA 26.10 00:00]; in the tunnel, with a delay on the obstruction of the view of the same hour [CLARIDEN 06.10 18:40].
  • The cause, and how it was known: thermal effect on the camera, according to the expert, on the housing temperature read in the metrics [BREITHORN 19.10 00:00].
  • What is concluded: not a movement of the structure, nothing to do; do not trust these points [BREITHORN 19.10 00:00].
  • Variants: a lasting remainder is added to the drift of the camera (see "The camera tilts little by little") [MICHIGAN 16.10 12:00].
  • Labels: BREITHORN 19.10 00:00, MICHIGAN 28.10 16:00, NEBRASKA 28.10 20:00, CLARIDEN 06.10 18:40, BREITHORN 26.10 04:00, NEBRASKA 26.10 00:00, and 16 others: MICHIGAN 6, BREITHORN 4, NEBRASKA 3, CLARIDEN 3.

A step of the camera that the compensation does not erase

  • What happens: the camera moves all at once; the compensation catches up almost all of it, and the rest passes into the curves of the targets.
  • What is seen: a small step of the pan and the rotation [NORDEND 01.09 15:30], or a step of the tilt that the pose keeps [SNCF01 15.10 05:15]; the raw positions of all the sensors, references included, shifted together all at once, not quite by the same amount [CLARIDEN 20.10 00:15, E-006]; several targets jump together and come back at the next measurement [NORDEND 01.09 15:30, E-006], or keep a jump that stays [CLARIDEN 20.10 00:15]; the references at their level [NORDEND 01.09 15:30], or their own corrections that change [SNCF01 15.10 05:15]; images: nothing in front of the targets, a blurred foliage [E-006].
  • Over time: one measurement; or a step that stays [SNCF01 15.10 05:15]; or a rotation that drops over a morning, in the middle of the daily cycle, and does not come back [WILDSTRUBEL 08.10 05:30].
  • The cause, and how it was known: not established; a jerk of the camera or its mount, deduced from the pose [SNCF01 15.10 05:15]; wind is possible [E-006].
  • What is concluded: not a movement of the structure; of one measurement, "very slight and insignificant", nothing to do [E-006]; camera to watch if it repeats [NORDEND 01.09 15:30]; a remainder that lasts, to watch on the mount side [SNCF01 15.10 05:15]. The size of the raw step does not make the event: by day, steps of the same size leave nothing [E-006].
  • Labels: NORDEND 01.09 15:30, SNCF01 15.10 05:15, CLARIDEN 20.10 00:15, WILDSTRUBEL 08.10 05:30; [E-006], two instants of NORDEND on 03.09, labeled Visibility by NORDEND 03.09 13:15 and NORDEND 03.09 16:30 (PR07 faint, rejected at the first): the data there show the common raw step of all the sensors of the event.

The camera tilts little by little

  • What happens: the camera nods forward little by little, over weeks, and the compensation of this movement is incomplete: it corrects some targets better than others [BREITHORN 03.10 00:00, according to the expert]; on two references, it passes into all the targets [MICHIGAN 09.10 12:00].
  • What is seen: the tilt increases, the pan and the rotation decrease, the scale drops [BREITHORN 03.10 00:00, MICHIGAN 09.10 12:00]; the targets slide together in the image, without moving relative to each other [MICHIGAN 09.10 12:00]; a few drift slowly in the same direction, the others do not: those that drift the most are much closer to the camera, according to the expert [BREITHORN 03.10 00:00]; elsewhere those at the top and on the right [NEBRASKA 03.10 00:00]. The scene may deform with the camera, in both directions [NEBRASKA 16.10 20:00 (drift)]; the positions of the targets and of the references, relative to each other, change more and more, the reprojection increases [SNCF01 04.10 00:00 (Displacement)]. error_value stable [BREITHORN 03.10 00:00]. The drift follows the jerks of the temperature, but at equal temperature the camera is more tilted at the end of the month: link not established [BREITHORN 16.10 20:00].
  • Over time: weeks; advance, stop and slight recoil, resumption, on the same dates on the cameras of the same building, according to the expert [BREITHORN 03.10 00:00, BREITHORN 13.10 04:00]; it slows down without stopping [BREITHORN 16.10 20:00]; after weeks without measurement, a single measurement shows the camera much more tilted [WILDSTRUBEL 31.10 20:00].
  • The cause, and how it was known: given by the expert: the camera nods forward and its compensation is incomplete [BREITHORN 03.10 00:00]; three cameras tilting together point to a cause common to the building rather than a defect of one camera [BREITHORN 03.10 00:00]. Elsewhere not established: incomplete compensation, a reference that moves, or the structure [NEBRASKA 16.10 20:00 (drift), SNCF01 04.10 00:00 (Displacement)].
  • What is concluded: the targets do not move relative to each other: not a movement of the structure; the tilt is to be watched [BREITHORN 03.10 00:00]. When the scene deforms with the camera: movement of the structure not established, to be watched [SNCF01 04.10 00:00 (Displacement)]. Look at the mounting of the camera [WILDSTRUBEL 31.10 20:00].
  • Variants: the recoil can be read on the curves [BREITHORN 13.10 04:00, MICHIGAN 12.10 16:00, NEBRASKA 13.10 04:00]; the remainder of the cold mornings is added to it [MICHIGAN 16.10 12:00].
  • Labels: BREITHORN 03.10 00:00, MICHIGAN 09.10 12:00, NEBRASKA 16.10 20:00 (drift), SNCF01 04.10 00:00 (Displacement), WILDSTRUBEL 31.10 20:00, BREITHORN 13.10 04:00, and 5 others: MICHIGAN 2, NEBRASKA 2, BREITHORN 1.

F. The structure has moved

The structure expands and contracts with the temperature

  • What happens: a structure that heats up and cools down moves, normally for its kind; large, it takes time to cool, and the movement is slow.
  • What is seen: targets, on the structure: they move a lot; the corr_y hollows out by day [JUNGFRAU 13.10 07:14]; this bridge rises as it cools [JUNGFRAU 11.10 18:05]; the corr_x of most of them rises in the afternoon [JUNGFRAU 02.10 06:45]; on cool nights, the two targets at the edges of the bridge drift apart in corr_x, in opposite directions, then come back [JUNGFRAU 19.10 20:00]. References, off the structure: very stable, in position as in error_value. error_value of the targets very stable: they stay well visible. Reprojection: a little higher, without it being significant. Camera, fixed to the bridge: it moves from the morning; well compensated on days with little sun [JUNGFRAU 30.09 07:30]. Temperatures: the components of the camera "do not clearly show the phenomenon" [JUNGFRAU 13.10 07:14]; the automatic labels nevertheless date the phases on the housing temperature: it does draw the daily cycle, but ahead of the bridge, which follows it with a delay (seen in the data of 11 to 14.10).
  • Over time: the cooling starts at nightfall and reverses at daybreak [JUNGFRAU 11.10 18:05]; the warming from sunrise, all day [JUNGFRAU 13.10 07:14]; weak on an overcast day [JUNGFRAU 07.10 06:30]; the cooling may stay flat until the middle of the night [JUNGFRAU 13.10 18:15].
  • The cause, and how the expert knew it: deduced: the targets on the structure move, the references off the structure do not, and the visibility does not change.
  • What is concluded: the bridge moves, which is normal for a structure of this kind. The stability of the error_value rather suggests a poor compensation by the camera [JUNGFRAU 11.10 18:05]; the movement is badly compensated by the camera fixed to the bridge [JUNGFRAU 13.10 07:14].
  • Variants: mixed with a drop in visibility, hard to read [JUNGFRAU 07.10 08:36]; with rain, the thermal part cannot be separated [JUNGFRAU 25.10 07:30]; under the loss of the references, the underlying trend stays this movement [JUNGFRAU 14.10 23:02].
  • Labels: JUNGFRAU 11.10 18:05, JUNGFRAU 13.10 07:14; JUNGFRAU 02.10 06:45, JUNGFRAU 30.09 07:30, JUNGFRAU 19.10 20:00, JUNGFRAU 25.10 07:30, JUNGFRAU 13.10 18:15, and 50 others: JUNGFRAU 50 (one station). The 24 h wave of thermal-cycle.md, noted "PS des Coudres", is this one.

A rise over several days with the drop in temperature

  • What happens: the temperature keeps dropping for days, which breaks the thermal cycles; the structure, which rises as it cools, rises all that time: a plausible and coherent explanation, not established; there may still be sun, hence thermal effects on the bridge.
  • What is seen: the clear phases fade all at once; the targets rise for several days, then stabilize; the corr_y rises almost without stopping for two nights and the day between them [JUNGFRAU 16.10 16:45]. Temperatures of the camera components: they fall and only keep dropping. Other small things are mixed in, such as a loss of visibility toward the end.
  • Over time: several days, mid-October.
  • The cause, and how the expert knew it: deduced: the rise follows the temperature; the other thermal labels show that the bridge rises when the temperature drops.
  • What is concluded: a real movement can reasonably be set aside; to understand these data, one must look at the temperature.
  • Labels: JUNGFRAU 16.10 14:18; JUNGFRAU 16.10 16:45 (one station).

A real movement of the structure

  • What happens: the structure really moves, and drags the camera and some references along.
  • What is seen: references: at night, they move relative to each other, not all of them (three shift in the image, the fourth, D-302-C, stays motionless); their coherence degrades clearly. Targets: their corrections break away without their position in the image moving, except around the motionless reference; in the morning, a group of neighbouring targets moves alone in the image; then all the sensors, references included, move together all at once, come back in part and stay shifted. The corrected displacements keep a step, larger the farther the target is from the references. A target that loses its reflector stays just under the threshold: its kept values do not measure the structure [E-004]. Images: nothing (no obstacle, no fogging, no change in lighting).
  • Over time: a night then a morning; the step stays more than a week and disappears after a short outage of the measurements.
  • The cause, and how the expert knew it: a tunnel under construction; the other measurement systems of the site confirm the movement. The AIMsight documentation reads the same episode as an instability of the reference network; the two do not exclude each other (barcelona-incident.md).
  • What is concluded: intervention, the site was stopped. The camera and the references having moved too, the measured amplitude is distorted; the true one is not established.
  • Labels: CLARIDEN 18.09 23:20 (one station); [E-004], the same night.

Targets move slowly, alone, then come back

  • What happens: one or a few targets change position slowly, alone, for hours or days, without the camera moving or the view being obstructed, then come back, partly or beyond; the cause is not established.
  • What is seen: a drift in small steps, then a new level; a row that rises then comes down again [CLARIDEN 06.10 22:00, CLARIDEN 11.10 06:00]; a single target that goes down night after night [WILDSTRUBEL 25.09 02:30]. Their positions in the image move away from the others in the direction of the curves and come back [CLARIDEN 06.10 22:00, WILDSTRUBEL 25.09 02:30]. Ordinary error_value, camera motionless, temperatures unrelated [CLARIDEN 06.10 22:00]; nothing in front of the targets [CLARIDEN 11.10 06:00]; workers near the camera, no link established [CLARIDEN 22.10 08:45].
  • Over time: hours to days; a slow return, partial [CLARIDEN 25.10 18:00] or beyond [CLARIDEN 22.10 17:00]; brief steps that stay [CLARIDEN 28.10 19:20].
  • The cause, and how it was known: not established, in a tunnel under construction [CLARIDEN 06.10 22:00]; a target already hindered by something at the same hour of the night [WILDSTRUBEL 25.09 02:30].
  • What is concluded: a movement of these targets is not excluded: to be checked with the other measurements of the site, watch the row [CLARIDEN 06.10 22:00].
  • Variants: a row that rises by day during works seen on the images and comes down in the evening [CLARIDEN 22.10 08:45]; an agitation in jerks, two references that drift slightly apart, then a shift that stays [CLARIDEN 24.10 18:20].
  • Labels: CLARIDEN 06.10 22:00, CLARIDEN 11.10 06:00, CLARIDEN 22.10 08:45, CLARIDEN 28.10 19:20, CLARIDEN 24.10 18:20, WILDSTRUBEL 25.09 02:30, and 5 others: CLARIDEN 5.

A target or a small group jumps and stays shifted

  • What happens: a target, or a few neighbours, move all at once to another level and stay there, without the camera moving or a cause being visible; they sometimes come back in one step.
  • What is seen: from one measurement to the next, the corr_x or the corr_y of a single target [FLORIDA 22.10 03:15, NORDEND 05.09 12:00], its raw position shifted by as much against the others [NEBRASKA 16.10 20:00 (COLJ19), WILDSTRUBEL 06.10 08:30]; or a group at the edge of the image together, one target rejected afterwards at almost every measurement [NIEDERHORN 30.10 01:00]. error_value a little lower [FLORIDA 22.10 03:15], barely higher [NORDEND 05.09 12:00], or changing level for the group [NIEDERHORN 30.10 01:00]. References and camera motionless; images: nothing [NORDEND 05.09 12:00].
  • Over time: all at once; the step lasts from a day [NEBRASKA 16.10 20:00 (COLJ19)] to more than a month [NORDEND 05.09 12:00]; it comes back in one step to its previous level [NORDEND 05.09 12:00, FLORIDA 15.10 04:00]; two steps in the same direction, the second in an outage [WILDSTRUBEL 06.10 08:30, WILDSTRUBEL 07.10 13:30].
  • The cause, and how it was known: not established: the spot perhaps changed look, or the target moved [FLORIDA 22.10 03:15, NIEDERHORN 30.10 01:00]; just after a rain [FLORIDA 09.10 08:00, E-014].
  • What is concluded: a movement of the target is not excluded: to be checked on its images and on site, to be watched [NORDEND 05.09 12:00, NEBRASKA 16.10 20:00 (COLJ19)].
  • Variants: the step that brings a target back to its level of a month earlier is a return, even labeled under another cause (a mark on the glass) [NORDEND 05.09 12:00, NORDEND 08.10 02:15].
  • Labels: NORDEND 05.09 12:00, FLORIDA 22.10 03:15, NEBRASKA 16.10 20:00 (COLJ19), WILDSTRUBEL 06.10 08:30, NIEDERHORN 30.10 01:00, FLORIDA 09.10 08:00, and 3 others: FLORIDA 1, NORDEND 1, WILDSTRUBEL 1; [E-014] (learned, not revalidated): e09, e10 and PWD_RMP_B16_026_C09 of NIEDERHORN on 02.10 in the evening, after the rain, the same targets as the group of NIEDERHORN 30.10 01:00; no label carries this first step (NIEDERHORN 02.10 20:00 is the train of the following measurement).

A step taken during a measurement outage

  • What happens: during a stop of the measurements, targets change level; on resumption, they do not find the previous one again.
  • What is seen: many targets resume at another level from the first measurement, the raw positions of a few shifted against the others [CLARIDEN 03.10 17:45, WILDSTRUBEL 09.09 21:45]; or a single neighbouring group, references intact [NIEDERHORN 15.10 22:00 (No data)]; the references slightly relative to each other, which changes the base [WILDSTRUBEL 09.09 21:45]; the error_value and the scale of the camera no longer the same [CLARIDEN 03.10 17:45]; a blurred object very close to the lens on resumption [CLARIDEN 03.10 17:45].
  • Over time: the step lasts weeks, the daily cycle on top [NIEDERHORN 15.10 22:00 (No data)]; a target comes back later near its level [WILDSTRUBEL 30.09 02:15].
  • The cause, and how it was known: not established: an intervention or a setting during the stop, or a real movement.
  • What is concluded: do not compare the levels before and after; to be checked with the operator; for a target shifted alone, a movement is not excluded [WILDSTRUBEL 09.09 21:45].
  • Labels: CLARIDEN 03.10 17:45, NIEDERHORN 15.10 22:00 (No data), WILDSTRUBEL 09.09 21:45, WILDSTRUBEL 30.09 02:15.

G. Station configuration

Targets that are twins of their references

  • What happens: the same prism serves as target and as reference, for lack of a possible reference in this part of the field, because the system does not yet accept a sensor with two roles.
  • What is seen: a single spot for the two sensors of each pair; their positions in the image move together, at a constant offset; the same error_value at the same instants; the corrections of the target stuck at zero, by day as by night, when an ordinary target has its own level and a much larger noise. When their common error_value passes the threshold, the twin loses its corrections and those of the reference, still written, leave zero [E-011, FLORIDA 18.10 05:00, seen in the data]; when the reference is rejected, its twin is too, or leaves its zero [FLORIDA 02.10 09:45, FLORIDA 09.10 09:45].
  • Over time: permanent, on all the measurements of the station.
  • The cause, and how the expert knew it: the configuration of the station, intended.
  • What is concluded: it is not a failure; nothing to check or to do. These targets are read as their reference: their flat curve does not prove that the structure is motionless at that place, and each episode that touches the reference counts one more sensor than in reality; in the labels, a twin is named with its reference [FLORIDA 08.10 03:15].
  • Labels: FLORIDA 17.10 23:15; FLORIDA 02.10 09:45, FLORIDA 09.10 09:45, FLORIDA 08.10 03:15, and 10 others: FLORIDA 10 (one station); [E-011], Sellette 3 and p1_r of FLORIDA from 16 to 25.10.

2. Reading rules

What holds beyond one case. The rules for placing a label (bounds, splitting, family) are in labeling/labeling-rules.md.

The expert's rules

  • A jump is judged against the neighbours. If it brings the target back toward them, it is a return to normal, and the artifact was before; if it takes it away from them, it is to be treated quite differently. [CLARIDEN 18.09 00:00, E-010]
  • A high error_value is not a problem in itself: it is its variations that count. [JUNGFRAU 14.10 23:02]
  • At least three references are needed; dropping a single one is enough to distort everything: a kind of triangulation. [NORDEND 04.09 10:59, JUNGFRAU 26.10 01:56]
  • The reprojection error says how many references hold: at zero, only two remain, very bad sign; absent, only one, and no measurement is possible. Except on a station built with two references, where it is always at zero. [JUNGFRAU 06.10 05:45, JUNGFRAU 14.10 23:02, JUNGFRAU 27.10 00:56; FLORIDA 01.10 20:45]
  • The position of the targets rests on the network of references: references badly seen, targets badly anchored. [JUNGFRAU 14.10 23:02]
  • A very stable error_value says the targets stay well visible: while they move, it rather suggests a poor compensation than an obstruction of the view. [JUNGFRAU 11.10 18:05]
  • The kind of structure says whether a thermal movement is possible: a bridge yes; a rock face, too massive, or a road, no: the movement is then that of the camera. [JUNGFRAU 11.10 18:05, NORDEND 25.09 16:02, WILDSTRUBEL 27.09 04:15]
  • The camera can move through a thermal effect at a different rhythm from the structure, not having the same heating time. [JUNGFRAU 11.10 18:05, JUNGFRAU 13.10 07:14]
  • The temperature that counts is that of the camera housing, not that of the air around, even sheltered. [expert, 2026-11-04, BREITHORN 19.10 00:00]
  • A change of regime is understood through the temperature: it must be looked at. [JUNGFRAU 16.10 14:18]
  • A tilt that rises means the camera nods forward. [expert, 2026-11-03, BREITHORN 03.10 00:00]
  • Times are in UTC, and the station can be anywhere: day and night are judged on the images. [WILDSTRUBEL 27.09 04:15]
  • A rejected measurement is removed from the display, not a measurement just under the threshold: a target left under the threshold keeps its jump. [NIEDERHORN 30.09 21:00, NIEDERHORN 01.10 11:00]

Those that the reviewed labels and the events establish

  • What moves in the image is read before the cause, on the raw positions against the overall movement (overview.py --rigid): the targets do not move relative to each other, it is the camera [BREITHORN 03.10 00:00]; targets move alone, a movement is not excluded [CLARIDEN 06.10 22:00]; nothing moves and the curves jump, it is the base or the configuration [WILDSTRUBEL 01.10 13:00, CLARIDEN 29.10 23:45].
  • A level is also judged against the past of the target, over weeks: a jump that brings it back to an old level is a return. [NORDEND 05.09 12:00, NORDEND 08.10 02:15]
  • A drift specific to a few targets that follows the camera, on the same dates, is the scene deforming with it, not a second movement; it is separate only if it has its own dates. [NEBRASKA 16.10 20:00 (drift), NEBRASKA 16.10 20:00 (COLJ19)]
  • The direction of a thermal movement is read on the housing temperature, not on the hour; when the scale of the camera follows it, the effect is on the camera. [NORDEND 01.10 10:00, CLARIDEN 06.10 18:40]
  • The pose of the camera is reliable only if the references hold: a reference rejected, dropped or sliding against the targets distorts it; a single hindered target does not. [FLORIDA 27.10 01:45]
  • A problem target is recognized by its own rejections, almost every day, also on overcast or rainy days; rejections that come only with a seen condition (sun, rain) are episodes of that condition. [NIEDERHORN 01.10 00:00 (e01), NORDEND 16.10 19:00; counter-examples: NORDEND 10.08 13:30, SNCF01 19.10 13:30]
  • On a recent camera, the flux says whether a target is hidden: cut, it is, even under its threshold, and its point is false; whole, nothing is in front; the error_value says a measurement is degrading, flux, sharpness and dominance say why. [LENZSPITZE 17.10 05:00; case: quality-metrics.md]
  • A gap is read with the other stations and the metrics: at the same minutes elsewhere, a cause common and outside the stations; synchronization, image or unsent entries during the gap, the station runs; at the end of a batch, a delay only if the others do not yet have their data; a drawn-down battery on return of a station on mains, the mains lost. [JUNGFRAU 15.10 22:45, NORDEND 22.08 16:30, WILDSTRUBEL 31.10 20:15, CLARIDEN 18.10 13:45; case: gaps.md]
  • Neighbouring cameras are read together: the same movement, the same loss or the same gap on the cameras of the same building have a common cause; what touches only one is on its line of sight. [MICHIGAN 09.10 08:00, BREITHORN 25.10 20:00]
  • Neither a binary inlier, nor sigma0 alone, nor the number of references is enough; coherent references are not fixed references. [CLARIDEN 18.09 23:20, E-004; case: measurement-chain.md]

3. False friends

Those that the cards do not already say plainly; the others are in "What is concluded".

  • A reprojection at zero is not a perfect measurement: only two references remain. [JUNGFRAU 26.10 01:56, JUNGFRAU 06.10 05:45]
  • Jumps of many targets, all in the same direction, are not a movement when they fall at the reference switches: raw positions without a step, daily cycle that continues. [NORDEND 28.07 20:45]
  • A target that slides slowly away from its neighbours is not moving when its error_value rises with the sliding and a step brings it back toward them. [CLARIDEN 18.09 00:00]
  • A rise over several days is not necessarily a real movement: it may follow a lasting drop in temperature. [JUNGFRAU 16.10 14:18]
  • Corrections stuck at zero are not a failure: a twin of a reference. [FLORIDA 17.10 23:15]
  • No signal does not mean no obstruction: only the images show the web. [NORDEND 03.09 18:01]
  • A web is not the spider: the web veils in part for hours or days; the spider hides everything for the length of one measurement. [NORDEND 31.07 20:15]
  • References that do not move in the rain do not contradict the rain: higher up, they suffer it less. [NIEDERHORN 01.10 04:00]
  • The step of a real movement is not its amplitude when the camera and the references have moved too. [CLARIDEN 18.09 23:20]
  • A swarm of insects is not a train: same signature; only the image tells them apart. [NIEDERHORN 13.10 22:00]
  • A daytime agitation is not turbulence when the error_value of the targets does not rise and the pose fidgets: it is the camera heating up. [NIEDERHORN 03.10 12:00, WILDSTRUBEL 27.09 04:15]
  • A drop in temperature at the same moment does not make a thermal effect: in the tunnel, the bump of a group comes with the error_value and returns before the scale of the camera. [CLARIDEN 06.10 18:00, CLARIDEN 06.10 18:40]
  • A drift that follows the temperature in trend is not thermal when, at equal temperature, the camera does not return to the same level. [BREITHORN 16.10 20:00, MICHIGAN 16.10 12:00]
  • The return of the error_value is not the end of an obstruction: after a mark on the glass, the curves stay shifted. [NORDEND 08.10 02:15]
  • Bigger spots on the image are not wider spots: when the light drops, the image makes halos of them. [LENZSPITZE 17.10 10:55]
  • A spot on a thumbnail that follows the measurement is not necessarily there: the frame follows a measurement perhaps latched onto the neighbour; only a fixed frame shows it. [FLORIDA 20.10 05:15]

4. The context of the stations

To read the unlabeled ranges of these stations; not a rule for the others. "Often": in many labels. Each event is attached to its station by its coded data, identical value for value to data/<STATION>/ at the same instants (labeling/auto/revue-croisee/2026-11-04/carte/outils/rattacher_events.py).

Station Structure Camera and references What comes back in its labels Events
BREITHORN in a building, with MICHIGAN and NEBRASKA [expert] one measurement every four hours; COLJ19 and POUI19 much closer to the camera [expert]; the camera nods forward over the month, incomplete compensation [expert] often, vehicles parked at night in front of the targets and references at the bottom (COLI22, COLI24, POUI22, POUI24, COLH20''', REFH22, REFPOU22) [BREITHORN 25.10 20:00]; the camera that moves on cold nights [BREITHORN 19.10 00:00]; the drift, common to the three cameras [BREITHORN 03.10 00:00] —
CLARIDEN a tunnel under construction, other measurement systems [CLARIDEN 18.09 23:20] four references, D-302-C, PR-291, PR-293, PR-295; PR-284, local reference read as a target; D-302-B and PR_BG on the same prism [CLARIDEN 29.10 23:45]; images always dark; on mains the obstacle in front of D-303-A [CLARIDEN 18.09 00:00]; the obstruction of the tunnel on the same group [CLARIDEN 06.10 18:00]; the scale at temperature drops [CLARIDEN 06.10 18:40]; rows B and C that go and come back [CLARIDEN 06.10 22:00]; P-300-B rejected since a setting; mains cuts [CLARIDEN 18.10 13:45] [E-004] (18.09, real movement), [E-010] (18.09, D-303-A)
FLORIDA — two references only, p1_r and r1_r, each with its twin (Sellette 3, Pont Ernen) [FLORIDA 17.10 23:15] the twins; r1_r that weakens at night and is lost in the morning [FLORIDA 22.10 11:15]; Sellette 6 at the threshold all month, Massif 2 that degrades [FLORIDA 06.10 10:15]; one-measurement gaps at night, at the same instants as at SNCF01 [E-011] (16 to 25.10, Sellette 3 and p1_r)
JUNGFRAU a bridge, which rises as it cools [JUNGFRAU 11.10 18:05] camera under the deck, fixed to the bridge; three references off the bridge, REF_D, REF_G, REF_CT the phases of the bridge almost every day [JUNGFRAU 13.10 07:14]; REF_G and REF_D badly seen or lost: a truck [JUNGFRAU 06.10 05:45], the sun behind REF_G [JUNGFRAU 12.10 15:45], nights [JUNGFRAU 26.10 01:56]; PM_G rejected at all measurements [JUNGFRAU 30.09 00:00]; BJ_G and BJ_D that drift apart on cool nights [JUNGFRAU 19.10 20:00] [E-001] (night of 13 to 14.10, REF_G and REF_D)
LENZSPITZE a railway seen from under a bridge, a site and machines next to it nine references, including the distant 8011 to 8014; recent camera, with image quality; x, y relative; only temperature_deg_c is filled the nights when the distant part goes dark [LENZSPITZE 16.10 22:25]; the light that drops on almost all the targets [LENZSPITZE 17.10 10:55]; pass-bys never seen, read on the flux [LENZSPITZE 17.10 05:00]; 92H hidden by the site [LENZSPITZE 30.10 13:50]; data since 16.10 —
MICHIGAN in the same building, a factory, indoors [expert] one measurement every four hours; two references since the loss of REFPOU22 on 09.10 [MICHIGAN 09.10 08:00]; COLH21' and REFH22' on the same point until 05.10 [MICHIGAN 03.10 00:00] the cold mornings [MICHIGAN 28.10 16:00]; the drift common to the building [MICHIGAN 16.10 12:00] —
NEBRASKA in the same building [expert] one measurement every four hours; two references since the loss of REFCOLH22, at the same measurement as MICHIGAN [NEBRASKA 09.10 08:00] the drift and the scene that deforms with it (POUH20, COLH21) [NEBRASKA 16.10 20:00 (drift)]; the cold nights [NEBRASKA 28.10 20:00] —
NIEDERHORN railway tracks references higher up, on the road above the tracks (PWD_REF08 to PWD_REF12); floodlight often, trains [NIEDERHORN 30.09 21:00]; rain at night [NIEDERHORN 01.10 04:00]; a plant in front of PWD_IMP_B07_016_C09 [NIEDERHORN 02.10 09:00]; the rising sun in the field [NIEDERHORN 03.10 12:00]; the thermal cycle of the camera [NIEDERHORN 13.10 13:00]; targets rejected always or at a fixed hour (e01, e03 to e12, PWD_RMP_B15_017_C09, PWD_IMP_B08_024_C09); insects [NIEDERHORN 13.10 22:00] [E-005] (30.09 to 03.10: trains, rain, plant); [E-013] (26.09, PWD_IMP_B08_024_C09); [E-014] (02.10 in the evening, e09, e10, PWD_RMP_B16_026_C09)
NORDEND quarry, rock face too massive to move with the temperature [NORDEND 25.09 16:02] three references, CTR02, CTR03, CTR04; CTR01, local reference read as a target; infrared lighting at night; DIFF images webs and spiders [NORDEND 28.07 20:45]; the thermal cycle of the camera, every day [NORDEND 02.10 06:00]; CTR01 in the sun in the afternoon [NORDEND 10.08 13:30], then problem target [NORDEND 16.10 19:00]; PR04 that degrades then recovers [NORDEND 17.08 05:15]; the spots at the left edge that go dark at night [NORDEND 01.10 03:00]; marks on the glass [NORDEND 25.10 05:45]; CTR04 dropped at times [NORDEND 04.09 10:59] [E-002] (28 to 31.07, web); [E-006] (03.09, step of the camera); [E-007] (01.09, one measurement less sharp); [E-008] (01 and 02.09, PR04)
PS_DES_COUDRES it is JUNGFRAU: the name of the export received from this station ("PS des Coudres", displacement-columns.md) the same sensors (REF_G, REF_D, REF_CT, BJ_G, PBJ_C, PM_C, PM_D); its label describes, word for word, the night of [E-001], whose data are those of data/JUNGFRAU/ PS_DES_COUDRES 13.10 23:45 (awaiting review) = JUNGFRAU 13.10 23:45 [E-001]
SNCF01 a masonry wall, along railway tracks three references, BG.3.1, R1, R2; no system metrics, hence no temperature often, trains, which sometimes hide two references [SNCF01 05.10 15:45]; daytime turbulence [SNCF01 04.10 08:15]; the thermal cycle without rest at night [SNCF01 05.10 06:30]; targets rejected always or at the threshold (BG.1, F4.1, CV.7, F4.4, F4.3, CV.6); the scene that deforms over the month [SNCF01 04.10 00:00 (Displacement)]; stops of the station alone [SNCF01 20.10 12:45] —
WILDSTRUBEL a road [WILDSTRUBEL 27.09 04:15], in a town six references, including R1 and R2 at the edge of the road; R3 invisible by day [WILDSTRUBEL 01.10 00:00]; solar battery often, pass-bys of vehicles and passers-by, which drop R1 and R2 together [WILDSTRUBEL 28.09 05:15]; parked cars [WILDSTRUBEL 05.09 03:30]; turbulence by day [WILDSTRUBEL 01.09 03:45]; the thermal cycle, well compensated [WILDSTRUBEL 01.09 03:30]; the station that shuts down at night for lack of energy [WILDSTRUBEL 01.10 03:15], and long stops [WILDSTRUBEL 09.10 12:30] [E-003] (01 to 04.09, turbulence and sun); [E-012] (26.09 in the morning, parked cars)

5. Differences with the event map

What the labels say differently from events/carte-des-evenements.md; the user decides.

  • Web, insects (detector visibility): a web can hide a reference without any signal [NORDEND 03.09 18:01]; insects hide everything for the length of one measurement, like a train [NIEDERHORN 13.10 22:00].
  • Pass-by ("references intact"): a truck, a train, a vehicle, the arm of a machine also hide references and make the base jump, sometimes until none is left [JUNGFRAU 06.10 05:45, SNCF01 05.10 15:45, LENZSPITZE 30.10 13:50].
  • Pass-by and "Deflection under a load" ("to be decided", [E-005]): the expert reads the jumps at the passage of a train as an effect of the passage: "the tracks do not move" [NIEDERHORN 30.09 21:00].
  • Object placed in front, too few references, a reference moved, camera knocked, dirty lens or glass, slow drift, a target or an area moved (hypothetical or without an event): seen [BREITHORN 25.10 20:00, MICHIGAN 09.10 08:00, WILDSTRUBEL 01.10 13:00, CLARIDEN 05.10 10:30, NORDEND 25.10 05:45, BREITHORN 03.10 00:00, NORDEND 05.09 12:00, NIEDERHORN 15.10 22:00 (No data)].
  • Rain, mist, fogging ("slight jump, negligible"): on two references, the rain cuts the pose [FLORIDA 01.10 20:45].
  • Lighting, lack of light: the sun on the background of a spot, at the same hours, makes a target rejected or the base jump [NORDEND 12.08 13:30, JUNGFRAU 12.10 15:45]; a faint spot is lost at the transition from night to day [FLORIDA 01.10 07:15].
  • Turbulence ("error_value higher by day"): also a fast noise of the corrected curves [SNCF01 04.10 08:15]; without a rising error_value, it is thermal [NIEDERHORN 03.10 12:00].
  • Masked references ("sigma0 rises"): the expert reads the reprojection, which falls to zero with two references and is missing with one [JUNGFRAU 14.10 23:02].
  • Thermal cycle of the camera ("compensated"): badly compensated, without rest at night, in jerks on cold nights, or on the scale alone without a daily cycle [NORDEND 25.09 16:02, SNCF01 05.10 06:30, BREITHORN 19.10 00:00, CLARIDEN 06.10 18:40].
  • Setting changed ("all the sensors at once"): also a single sensor or a pair [CLARIDEN 01.10 00:00 (P-300-B), MICHIGAN 03.10 00:00].
  • Dirty or dull target: rejected always, at a fixed hour, or from a date [NIEDERHORN 01.10 00:00 (e01), NORDEND 16.10 19:00 (CTR01)]; the twin is not labeled for itself [FLORIDA 17.10 23:15].
  • Data gap: common gap, energy, mains, station that runs during the gap are not in it [JUNGFRAU 15.10 22:45, WILDSTRUBEL 01.10 03:15, CLARIDEN 18.10 13:45, NORDEND 22.08 16:30].
  • Expansion of the structure ("rise by day, descent at night"): the direction depends on the structure, this bridge rises as it cools [JUNGFRAU 11.10 18:05]; seasonal movement (hypothetical): the closest [JUNGFRAU 16.10 14:18].

6. What the labels do not say yet

  • Causes not established: the drop in visibility over the whole scene [JUNGFRAU 07.10 08:36]; what veils the references at night [JUNGFRAU 26.10 01:56, E-001]; the nature of the obstacle [CLARIDEN 18.09 00:00]; the true amplitude of the movement [CLARIDEN 18.09 23:20]; the obstruction and the moving rows of the tunnel [CLARIDEN 06.10 18:00, CLARIDEN 06.10 22:00]; the drift and the loss of references of the building [BREITHORN 03.10 00:00, MICHIGAN 09.10 08:00]; the scene that deforms with the camera [SNCF01 04.10 00:00 (Displacement)]; the base that changes without a reference lost [WILDSTRUBEL 01.10 13:00]; the targets always rejected [NIEDERHORN 01.10 00:00 (e03)]; the lasting jumps of single targets [NIEDERHORN 30.10 01:00, E-014]; the gaps and the stops [JUNGFRAU 15.10 22:45, SNCF01 20.10 12:45].
  • Questions for the expert, opened by the data:
    • PS_DES_COUDRES is JUNGFRAU (section 4): its only label, awaiting review, is the night of [E-001] and of JUNGFRAU 13.10 23:45; it stays in labeling/manual/ (the user, 2026-11-05), and the automatic label JUNGFRAU 13.10 23:45 took its conclusion: small shifts, negligible, nothing to do [E-001].
    • NORDEND 03.09 18:01 says "no measurement is rejected": the data show PR07 and PR04 rejected for a few measurements around 23:00, PR07 under a thread of web according to NORDEND 03.09 22:45. No conflict: the expert's sentence is about the evening when the web hides CTR04; the rejection of PR07 comes later, under another thread. NORDEND 03.09 22:45 now says, according to the expert, that the reference hidden without a signal can silently distort values.
    • heat haze does not exist under a bridge (measurement-chain.md): LENZSPITZE, a track seen from under a bridge, has sunny days when the image trembles without losing light [LENZSPITZE 16.10 11:00]; decided (2026-11-05): the images show the track in full sun beyond the bridge, its shadow in the foreground only; it is the air above the sunlit track, not from under the bridge, and the turbulence labels of LENZSPITZE say so.
    • the cause of JUNGFRAU 07.10 08:36: wind (expert) or rain or mist (the automatic label JUNGFRAU 07.10 08:30); the images show an overcast day, the road wet at 16:00, without deciding: the automatic label keeps both hypotheses, cause not established.
    • the targets at the bottom of WILDSTRUBEL (PB1_2, PB2_2) in the morning: the cycle of the turbulence and the light, stronger on them [E-003], or a remainder of the camera [WILDSTRUBEL 24.09 04:00]? Both hold: the thermal effect is well compensated (the expert; the block of the targets follows the camera closely), with a small remainder at the bottom of the image, which the labels say so.
    • [E-012] says the hidden reference "not rejected": in the data, its error_value is above the threshold (so rejected) and its corrections stay written; [E-014] is carried by no label.
    • left over from the reviews: an intervention in the building on 09.10 in the morning, and is REFPOU22 of BREITHORN the lost prism of MICHIGAN [MICHIGAN 09.10 08:00]; the morning of 31.10 on MICHIGAN and NEBRASKA alone; a problem reference that does not make the targets jump [WILDSTRUBEL 01.10 00:00]; the direction of y when the camera nods forward [WILDSTRUBEL 08.10 05:30]; the edges of the bridge that drift apart at night [JUNGFRAU 19.10 20:00]; no reference kept while two remain well seen [LENZSPITZE 26.10 00:00].
  • Seen on a single station: the cards marked "(one station)" (energy, mains, web, glass, plant, twins, bridge, rise over several days, real movement); and, in other cards, the station with two references by construction (FLORIDA), no reference kept and the pass-by read on the flux (LENZSPITZE), the obstruction and the rows of the tunnel (CLARIDEN), the cold nights and the drift of the three cameras of one building.
  • Families of the event map: C has only degrading or badly set targets, none knocked or replaced; F has only one established real movement [CLARIDEN 18.09 23:20] and the thermal bridge.
  • Unlabeled ranges: before the first batch of each station (NORDEND before 10.08, except July; WILDSTRUBEL before 01.09; the others before the end of September or October; NIEDERHORN on 26.09 of [E-013]) and after 01.11.
  • Leftovers of the cross-review of 2026-11-04: the rise of the corr_x of PR12, PR13, PR15 and PR16 on 17.10 in the afternoon [NORDEND 17.10 12:45]; the jump of PR17 of 05.09 and the step of October that brings it back, to be reconciled [NORDEND 05.09 12:00, NORDEND 08.10 02:15]; the end of the NEBRASKA drift early in the month [NEBRASKA 03.10 00:00].
  • Awaiting review: PS_DES_COUDRES 13.10 23:45 (two references veiled at night; the station is JUNGFRAU); the learned events E-012, E-013, E-014, awaiting /review-event.