Markerless Lower Limb Biomechanical Data Extraction in Gait Analysis Using MediaPipe Pose

Biomechanical gait analysis is essential for evaluating motor performance and guiding rehabilitation; however, classical motion-capture systems are costly and require specialized infrastructure, severely limiting accessibility in routine clinical settings. This study presents and validates a markerless gait analysis system based on MediaPipe Pose, using a standard smartphone camera. Ten healthy male volunteers (aged ۲۰–۳۰ years) participated; three videos were recorded per participant across four real-world environments. An integrated Python software system was developed, comprising automatic landmark extraction, anthropometric calibration, gait cycle detection, and knee angle computation via the dot-product vector method, with output delivered through a graphical user interface and a structured PDF report. The system extracted walking speed (۱.۴۴ ± ۰.۰۹ m/s), stride time (۱.۰۸ ± ۰.۰۴ s), stride length (۱.۵۸ ± ۰.۰۸ m), cadence (۱۱۲.۴ ± ۵.۸ steps/min), and knee range of motion (۶۵.۹ ± ۱.۹° right; ۶۷.۶ ± ۴.۱° left). Stance and swing phase ratios (۶۱.۱% and ۳۸.۹%) closely matched normative references. Validation against Kinovea yielded a knee angle mean absolute error of ۴.۲–۴.۶° and a stride time error of ۰.۰۲۵ s, results comparable to published pose-estimation gait studies, supporting the system's suitability as a low-cost screening tool where marker-based systems are unavailable.

کلیدواژه ها:

MediaPipe Pose ، Gait Analysis ، Markerless ، Biomechanics ، Programming ، Python