ArMo: An Articulated Mesh Approach for Mouse 3D Reconstruction

Area of Science:

• Neuroscience
• Biomechanical Engineering
• Computer Vision

Background:

• Accurate 3D animal pose estimation is crucial for behavioral analysis in neuroscience research.
• Current keypoint tracking methods offer limited pose information, representing animals as sparse points.
• 3D mesh modeling can provide a more comprehensive pose representation but hasn't been applied to mouse behavior.

Approach:

• Developed ArMo, an articulated 3D mesh model specifically for laboratory mice.
• Applied ArMo to multi-camera recordings of head-fixed mice running on a spherical treadmill.
• Utilized an end-to-end gradient-based optimization to fit the 3D model to keypoint and point cloud data.

Key Points:

• ArMo reconstructs the 3D surface shape and pose of mice from video data.
• The model captures detailed animal surface shape, unlike sparse keypoint methods.
• Movement is compactly summarized as a time series of 3D skeletal joint angles.

Conclusions:

• ArMo provides a novel and comprehensive method for 3D mouse pose reconstruction.
• This approach offers a significant advancement over sparse pose representations in neuroscience.
• ArMo enables richer analysis of mouse behavior in research settings.