Spatiotemporal gait analysis using video from a moving observer.

Abstract

Spatiotemporal gait analysis is widely used in clinical and research settings to assess human movement. Diverse methods exist to obtain such data, but simple and versatile solutions suitable for real-world assessment are missing. This study aimed to develop and evaluate a method to perform spatiotemporal gait analysis using data collected by smart glasses worn by an observer accompanying and looking at a walking subject. The method, combining monocular 3D pose estimation and visual-inertial odometry, was assessed based on 608 walking sequences in varied indoor and outdoor conditions (54 healthy participants, 4.4 km in total). Compared to an inertial and video-based motion capture reference system, the proposed method detected 96.6% of heel-strike and toe-off events with a mean (± standard deviation) error of 8.7 ± 37.3 ms. Moderate to excellent agreements were found for temporal parameters (0.74 ≤ ICC ≤ 0.98), as well as for spatial parameters (0.87 ≤ ICC ≤ 0.99). Performance was generally consistent across environments, slopes, trajectories, viewpoints, and viewing distances, with differences of small or moderate effect sizes observed only in a few particular situations. Together, this study demonstrated that observer-based gait analysis can provide accurate and reliable spatiotemporal parameters in real-world conditions, offering a simple and scalable alternative.