Detect+Track: robust and flexible software tools for improved tracking and behavioural analysis of fish.

Area of Science:

• Ethology and Animal Behavior
• Computer Vision and Image Processing
• Bio-inspired Robotics and Navigation

Background:

• Accurate tracking of animal movement is crucial for understanding behavior and decision-making.
• Existing tracking methods often struggle with challenges like occlusion, variable lighting, and complex environments.
• There is a need for robust, adaptable, and generalizable tracking solutions for large-scale behavioral studies.

Purpose of the Study:

• To introduce Detect+Track, a novel video processing method for enhanced animal tracking.
• To improve the accuracy and robustness of tracking in challenging experimental conditions.
• To enable detailed analysis of movement and decision-making behaviors, such as gap selection.

Main Methods:

• Combines a deep learning-based object detector with a template-based object-agnostic tracker.
• Utilizes Voronoi tessellation and planar homology for virtual gate computation.
• Employs optical flow for estimating fish speed and movement direction.

Main Results:

• Accurate localization of fish centroids in challenging conditions (occlusion, variable lighting, body deformation, ripples).
• Detailed analysis of gap selection behavior in Picasso triggerfish (Rhinecanthus aculeatus).
• Precise estimation of fish speed and movement direction using optical flow.

Conclusions:

• Detect+Track offers a flexible and generalizable solution for animal tracking in complex environments.
• The method addresses key limitations of existing tracking tools, improving behavioral analysis.
• Provided data and code facilitate reproducibility and future innovations in behavioral tracking.