Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies

Area of Science:

• Ethology and behavioral neuroscience
• Computational biology and machine learning

Background:

• Accurate multi-animal pose tracking is crucial for studying social behavior in rodents under natural conditions.
• Existing markerless tracking methods struggle with occlusion and crowding, especially with indistinguishable individuals.

Purpose of the Study:

• To develop a novel method, virtual marker tracking (vmTracking), to enhance the accuracy of multi-animal pose tracking.
• To address the limitations of current tracking tools in challenging social interaction scenarios.

Main Methods:

• Introduced virtual markers to maintain individual animal identity across video frames.
• Developed a protocol for assigning virtual markers and tracking animals in labeled videos.
• Showcased vmTracking's applicability to existing markerless tracking data through additional processing.

Main Results:

• Significantly improved multi-animal pose tracking accuracy, particularly in crowded and occluded conditions.
• Successfully maintained individual animal identity throughout the tracking process.
• Provided a reliable foundation for quantitative social interaction analysis.

Conclusions:

• Virtual marker tracking (vmTracking) offers a robust solution for accurate multi-animal pose tracking in complex social environments.
• This method facilitates more reliable quantitative analysis of rodent social behavior under semi-natural conditions.