3D Gait Analysis for the Elderly Mobility Based on Multiple RGB Cameras

Walking is a fundamental movement that undergoes performance decline in the elderly. Traditional gait analysis relies on professional motion capture cameras and reflective markers, which are very expensive and cumbersome to use. Recent advancements in human pose estimation and deep learning have shown the feasibility of markerless motion analysis using regular RGB cameras. However, these methods may have deviations due to factors like light conditions, camera views, subject stature, and lack of biomechanical constraints. While some studies have compared markerless systems to professional motion capture systems, few have examined if these systems can still distinguish different motion patterns in real-world applications despite detection errors. This study presents a gait analysis system utilizing four RGB cameras and prior biomechanical knowledge to measure gait patterns in both elderly and young subjects. Personalized skeletal models were created for each subject on OpenSim, driven by 3D human keypoints obtained through human tracking, pose estimation and triangulation algorithms. Gait spatial temporal parameters and kinematics were then calculated for validation. When compared to the Vicon motion capture system, the Pearson correlation coefficients for the elderly gait were found to be 0.97 for hip flexion/extension, 0.92 for hip adduction/extension, 0.98 for knee flexion/extension and 0.85 for ankle dorsi/plantar flexion. Comparisons between the elderly and young individuals revealed shorter step and stride lengths and reduced range of motion in the elderly, which were consistent with existing literature, indicating the weaker gait performance in this population. These findings demonstrate that our RGB camera-based gait analysis system can effectively differentiate between different gait patterns. Therefore, in certain scenarios, motion analysis can be performed simply and affordable with our system.