Automated Gait Assessment for Rehabilitation Training Using Pose Tracking and Dynamic Time Warping

Background: In rehabilitation medicine, efficient gait analysis is crucial for evaluating postoperative recovery and frailty, especially given the increasing burden on clinicians due to an aging population. Objectives: This study aims to conduct preliminary validation of an automated linear walking evaluation system using 2D AI posture tracking. By evaluating the basic accuracy of the system on healthy individuals, we aim to establish a technical foundation for future introduction into clinical rehabilitation settings. Methods: In this observational study, we utilized a standard visible light camera for practical use. To evaluate accuracy, we compared 2D AI tracking against a gold-standard three-dimensional (3D) motion capture system during normal walking trials with 10 healthy participants. Specifically, we employed Dynamic Time Warping (DTW) to temporally align the asynchronous data streams from the 2D and 3D systems, ensuring precise comparison of joint angles. Results: Following the DTW-based alignment, the similarity with the 3D system was 0.806 ± 0.094 overall (Left: 0.797 ± 0.101, Right: 0.814 ± 0.086). Conclusions: In this preliminary validation, the proposed 2D AI posture tracking showed good agreement with the gold standard 3D motion capture for gait in healthy individuals. While the average systematic bias was within clinically acceptable limits, the observed limits of agreement suggest that this system is currently optimal as a foundational tool for gait screening. These results establish a technical foundation for the clinical application of this system.