AI-assisted digital video analysis reveals changes in gait among three-day event horses during competition.

Abstract

The value and welfare of performance horses is closely tied to locomotor behaviors, but we lack objective and quantitative measures for these characteristics, and qualitative approaches for assessing gait do not provide measures suitable for large-scale biomechanical research studies. Digital video analysis utilizing artificial intelligence-based strategies holds promise to meet the need for an economical, accurate, repeatable and objective technique for field quantification of equine locomotion. Here we describe pilot work using a consumer-level digital video camera to capture high-resolution and high-speed videos of horses moving at the trot during mandatory inspections for international-level eventing competitions. We assessed 194 horses from five different competition venues, recorded at pre-competition (first) and post-cross-country (second) inspections as a model of gait change following exertion. We labeled twenty-six keypoints on each frame with DeepLabCut and processed the resulting tracking data using MatLab to derive quantitative gait parameters. Once trained, the DeepLabCut model labeled the 388 videos in just minutes, a task that would have otherwise taken months of human effort to complete. A Generalized Linear Mixed Model (GLMM) examining seven gait parameters identified significant changes in duty factor, speed, and forelimb swing range following the completion of the cross-country phase (P ≤ 0.05). Despite some limitations, video analysis through artificial intelligence proved capable of quantifying several gait parameters quickly, efficiently, and without the need for specialized equipment, making this tool a promising option for future biomechanical research in the athletic horse.