Detecting and Tracking of Multiple Mice Using Part Proposal Networks.

Area of Science:

• Neuroscience
• Computer Vision
• Animal Behavior

Background:

• Automated quantification of mouse social behaviors is crucial for neuroscience research.
• Current methods for tracking multiple mice often rely on intrusive artificial markers, hindering natural movement.
• Accurate object tracking is essential for analyzing complex mouse interactions.

Purpose of the Study:

• To develop a novel, non-intrusive method for continuously tracking multiple mice and their individual parts.
• To address the challenge of automated behavior quantification in neuroscience by improving object tracking accuracy.
• To introduce a new dataset for evaluating multi-mouse part detection and tracking.

Main Methods:

• A deep-learning-based scheme for efficient and robust detection of mouse body parts.
• A Bayesian-inference integer linear programming (BILP) model for joint assignment and association of detected parts.
• Introduction of the Multi-Mice PartsTrack dataset for quantitative evaluation.

Main Results:

• The proposed method accurately tracks multiple mice and their parts without artificial tagging.
• The novel approach outperforms existing state-of-the-art methods in accuracy for mouse behavior analysis.
• Demonstrated generalization capabilities on tracking other species like zebras and locusts.

Conclusions:

• The developed non-intrusive tracking method significantly advances automated analysis of mouse social behaviors.
• The Multi-Mice PartsTrack dataset provides a valuable resource for future research in multi-mouse tracking.
• This work offers a robust solution for accurate and generalized animal behavior quantification.