- base_connaissances/ → snippets/ (corpus/ → textes/, instructions.md → consignes_redaction.md, decoupage.py → fichiers.py, importer.py → envoyer.py) ; python3 -m snippets.envoyer. - instance-de-test/ : 1_demarrer.sh, 2_creer_boite_et_cle.py, 3_brancher_ia.py, 4_essayer_brouillons.py, voir_empreintes.sh, connexion_admin.py. - ollama/verifier.py → ollama/verifier_ollama.py ; tests, tâches VS Code et docs alignés. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
30 lines
1.9 KiB
Markdown
30 lines
1.9 KiB
Markdown
# Where should reference targets be placed?
|
|
|
|
Reference targets are used to estimate and compensate for movements of the camera itself.
|
|
Their placement therefore has a direct influence on the quality of this compensation.
|
|
Whenever possible, references should be installed on elements that can reasonably be
|
|
considered stable and distributed across the image, ideally surrounding or framing the area
|
|
containing the measurement points. A good horizontal and vertical distribution makes it easier to
|
|
distinguish camera pan, tilt and roll and avoids extrapolating the correction from references
|
|
concentrated in a small part of the image.
|
|
It is also preferable to place the references at a distance comparable to or greater than that of
|
|
the measurement points, ideally in the background. More distant references are less sensitive to
|
|
parallax effects caused by small translational movements of the camera support.
|
|
Conversely, references located significantly in the foreground may show a relatively large
|
|
apparent displacement following a small lateral movement of the camera. If this information is
|
|
then used to correct points located much farther away, the camera movement may be
|
|
overcompensated.
|
|
In practice, preference should therefore be given to:
|
|
• supports whose stability can reasonably be assumed;
|
|
• references located as far from the camera as practical, preferably in the background of the
|
|
monitored area;
|
|
• references distributed around the measurement area rather than grouped in a single part of
|
|
the image;
|
|
• references that will remain reliably visible over time, with limited risk of obstruction by
|
|
vegetation, vehicles, equipment or other objects.
|
|
If no perfectly fixed references are available, several measurement points can also be used
|
|
together as a reference network. Camera movement is then estimated robustly from the overall
|
|
network, while the system remains sensitive to differential movements between individual points.
|
|
|
|
Source : AIMsight FAQ, rév. 1.5
|