Evaluation of the Clinical Utility of a Gait Analysis System Using Pose Estimation Techniques in Physical Therapy

Conventional observational motion analysis in physical therapy is subject to biases that require a more quantitative approach to gait analysis. This study presents the development of a gait analysis system using pose estimation techniques and explores the usefulness of quantitative gait analysis in clinical rehabilitation. The proposed system employs a PC and a webcam to identify leg joint angles by estimating keypoints using OpenPose. Initially, the system was validated in a laboratory against 3D position measuring equipment, demonstrating promising accuracy, particularly in the estimation of knee joint angles. Next, we examined whether the accuracy of the gait analysis verified in the laboratory was adequate for actual rehabilitation. The validation results revealed a statistically significant difference between the healthy group and the left and right hemiplegic patient groups. The machine learning classification achieved a three-class classification between healthy subjects' hemiplegic and legs with hemiplegic disease with an accuracy of 80%. The proposed simple motion analysis system, using posture estimation technique contributes to the quantitative evaluation of gait analysis in physical therapy.