A Markerless Approach for Full-Body Biomechanics of Horses

Area of Science:

• Veterinary Medicine
• Biomechanics
• Computer Vision

Background:

• Quantifying equine kinematics is crucial for clinical evaluation, research, and performance analysis.
• Existing methods for equine motion capture are often complex and labor-intensive, limiting their widespread adoption.
• There is a need for more accessible and efficient techniques to analyze horse movement.

Purpose of the Study:

• To develop and validate a novel markerless motion capture methodology for calculating three-dimensional, full-body equine kinematics.
• To offer a scalable and labor-efficient alternative to traditional, instrumented motion capture techniques.
• To assess the accuracy and potential of this new approach for equine kinematic analysis.

Main Methods:

• A markerless motion capture system was developed using multi-camera video data.
• Neural networks were employed to identify and triangulate skeletal landmarks (markers) in 3D.
• An equine biomechanics model was scaled, and inverse kinematics (IK) was used to generate kinematic trajectories.
• The methodology was tested on a horse across three gaits, with various neural networks evaluated.

Main Results:

• The markerless system successfully calculated full-body equine kinematics without requiring instrumentation on the animal.
• Neural networks predicted over 78% of markers within 25% of the radius bone length.
• Root-mean-square-error for joint angles was less than 10 degrees for most degrees of freedom, demonstrating high accuracy.
• Training neural networks on diverse datasets improved accuracy and curve similarity in joint angle predictions.

Conclusions:

• The developed markerless motion capture methodology shows significant potential for accurate and efficient full-body equine kinematic analysis.
• This approach offers a scalable and less labor-intensive alternative to traditional methods, facilitating broader application in research and clinical settings.
• The study validates the use of neural networks and multi-camera video for advancing equine biomechanics research.