GroupTracker: Video tracking system for multiple animals under severe occlusion.

Area of Science:

• Ethology
• Neuroscience
• Computer Vision

Background:

• Understanding complex animal behaviors requires analyzing interactions of multiple individuals.
• Accurate tracking of individuals, especially during positional overlap (occlusion), is crucial for behavioral analysis.

Purpose of the Study:

• To develop a robust multiple animal tracking system capable of handling severe occlusion.
• To enable quantitative analysis of high-order nervous system functions through detailed behavioral studies.

Main Methods:

• Developed GroupTracker, a novel multiple animal tracking system.
• Implemented an expectation-maximization scheme with eigenvalue constraints for Gaussian mixture models to handle component overlap.
• Utilized 3D video tracking to monitor freely swimming medaka (Oryzias latipes).

Main Results:

• GroupTracker accurately tracks individual animals even under severe positional overlap.
• The system successfully monitored multiple medaka fish in a 3D environment with frequent interactions.
• Demonstrated the system's efficacy in resolving complex, overlapping movements.

Conclusions:

• GroupTracker provides an accurate solution for multiple animal tracking, even with significant occlusion.
• This system will facilitate the discovery of novel structures and patterns in animal interactions.
• Contributes to advancing the quantitative analysis of animal behavior and neural mechanisms.