Automated canine gait analysis using computer vision has the potential to extend objective canine gait assessment beyond specialized, controlled laboratories into domestic environments, but the field is comparatively less mature than human methods. This review explores the state-of-the-art for vision-based canine gait analysis, with a particular emphasis on single-camera (monocular) articulated pose and shape reconstruction, along with the extraction and interpretation of clinically relevant gait parameters. Across the literature, current pipelines reconstruct anatomical and surface representations of canines from images and video, yet rarely achieve the biomechanical fidelity or validation against gold-standard references such as motion capture, pressure walkways or fluoroscopy. Three requirements emerge from the literature: robust monocular 3D reconstruction sufficiently accurate to measure soft-tissue artifacts (approximately 10–20 mm), a standardized set of gait parameters aligned to veterinary assessment practices, and a shift from parts-based to holistic gait analysis. We highlight priority research directions to improve monitoring of canine gait in real-world settings including breed-representative datasets, synthetic training data to real-world data adaption, and ensemble learning for pathology identification. Addressing these gaps could allow for objective, longitudinal monitoring of canine gait in both veterinary practices and domestic environments. We advocate for increased interdisciplinary collaboration to foster innovation and establish new standards in the field.
Phelan et al. (Tue,) studied this question.