Systematic Literature Review on Methodologies for Evaluating the Accuracy of Vision-Based Markerless 3D Motion Capture Systems

This study systematically reviewed methodologies for validating the accuracy of vison-based markerless 3D motion capture systems and proposed reliable evaluation criteria. A systematic review protocol was implemented, and articles were retrieved from Web of Science, PubMed, ScienceDirect, and IEEE Xplore. After removing duplicate and irrelevant studies, 13 articles were analyzed. The review identified key experimental tasks, kinematic variables, and evaluation metrics. Monocular systems were primarily applied to simple, slow movements like walking and jumping, using basic variables such as joint angles and joint center positions. In contrast, multi-view systems demonstrated higher accuracy for complex, rapid movements like running, incorporating variables such as joint moments, ROM, and segment angles. Monocular systems commonly used MAE, while multi-view systems combined RMSE, ICC, and Bland-Altman analysis for validation. These findings provide a guideline for selecting methodologies to validate vision-based markerless 3D motion capture systems, tailored to monocular and multi-view systems.