Anipose

Anipose performs markerless 3D pose estimation of animal behavior from multi-camera video to quantify three-dimensional movement.

Key Features:

• Integration with DeepLabCut: Uses DeepLabCut for 2D keypoint detection as input to 3D reconstruction.
• 3D calibration module: Calibrates and aligns multiple camera views for accurate spatial reconstruction.
• Error-resolution filters: Implements filters to detect and correct errors in 2D tracking data.
• Triangulation with temporal and spatial regularization: Reconstructs 3D keypoints by triangulation that incorporates both temporal and spatial regularization.
• Efficient video processing pipeline: Processes large volumes of multi-view video for batch 3D pose estimation.
• Markerless 3D tracking: Estimates 3D poses without the use of physical markers.

Scientific Applications:

• System calibration evaluation (moving calibration board): Tests system accuracy and spatial reconstruction using a moving calibration board.
• Insect locomotion (fruit flies): Analyzes kinematics of leg movements in fruit flies walking on a treadmill, including joint rotation dynamics.
• Rodent motor control (mice reaching for pellets): Quantifies reaching kinematics in mice to study mammalian motor control.
• Human action analysis: Applies 3D pose estimation to humans performing various actions to study movement biomechanics.

Methodology:

Captures 2D keypoints using DeepLabCut, calibrates multiple cameras, triangulates keypoints across views with temporal and spatial regularization, applies error-correction filters, and processes multi-view video through a batch pipeline.