vmTracking enables highly accurate multi-animal pose tracking in crowded environments.

Area of Science:

• Ethology
• Biomedical Engineering
• Computer Vision

Background:

• Accurate multi-animal tracking is essential for behavioral analysis.
• Occlusion and crowding present significant challenges in pose tracking.
• Existing methods struggle with complex interactions in dense animal groups.

Purpose of the Study:

• To introduce virtual marker tracking (vmTracking) for robust individual identification.
• To overcome limitations of traditional markerless tracking in crowded scenarios.
• To enhance the precision and reliability of behavioral analysis in complex settings.

Main Methods:

• Developed vmTracking using virtual markers derived from markerless tools (maDLC, SLEAP).
• Applied virtual markers as cues for single-animal pose estimation (saDLC, SLEAP).
• Validated vmTracking across diverse species including mice, fish, and human dancers.

Main Results:

• vmTracking enabled consistent individual identification despite occlusion and crowding.
• Reduced the need for manual corrections and extensive annotation frames.
• Demonstrated broad applicability and variability across different animal models and human motion.

Conclusions:

• vmTracking offers a simpler, more effective solution for multi-animal pose tracking.
• Enhances the precision and reliability of analyzing complex naturalistic and social behaviors.
• Provides a valuable tool for advancing ethological and biomechanical research.