Knowledge Base

Problem

During a Novel Object Location or Novel Object Recognition test, EthoVision incorrectly identifies a stationary in-arena object as the animal's nose point when the animal is nearby but not interacting with it. The nose point 'sticks' to the object rather than tracking the actual animal nose.

Details

Occurs with 3-point (nose/center/tail) body-point tracking using Deep Learning detection in EthoVision XT. Stationary objects with edges or features that resemble body landmarks can confuse the model, particularly when the animal is in close proximity. The problem reduces tracking accuracy and may inflate zone contact or interaction metrics near the object.

Solution

1. Open Detection Settings and navigate to the Deep Learning body-point configuration.
2. Use the automated optimization/training tool within the Deep Learning detection panel to refine the model on representative frames from your video, including frames where the animal is near the object.
3. Review the detection output after optimization — the 'sticking' of the nose point to the object should occur less frequently.
4. If the problem persists, consider using the 'Exclude Object' or doughnut/cutout method in Arena Settings to mask the object area from detection.
5. Ensure lighting is even and the object does not cast shadows that alter its appearance to the detection algorithm.
6. As an additional measure, define a zone around each object and use the zone contact variable rather than raw nose-point position to measure object interaction, which is more robust to occasional misdetection.