Suggestion for Camera Location in Monocular Markerless 3D Motion Capture System: Focused on Accuracy Comparison With Marker-Based System for Upper Limb Joints

Abstract

A markerless motion capture system that analyzes human movements without markers has emerged. However, most studies use 2D coordinates to track basic movements from fixed viewpoints, with limited research on joint movement accuracy based on camera angles. This study evaluates a monocular markerless 3D motion capture system’s accuracy from various RGB camera positions and identifies optimal placements by comparing it with a marker-based system. Ten male participants in their 20s performed clinical range of motion (ROM) tasks, and joint angles were measured from different camera positions. The mean absolute error (MAE) and Pearson correlation coefficient between the two systems were analyzed. The neck joint showed the highest accuracy at 0°, while 45° and 315° were viable for flexion/extension. The shoulder joint performed best at 135°, 180°, and 315°, while the elbow joint showed the highest accuracy at 90° and 135°. The wrist joint had the lowest accuracy, but considering its correlation with flexion/extension, the 0°, 90°, and 270° positions may still be viable. For comprehensive upper body analysis, positioning the camera at 45° or 315° ensures accurate neck assessment while enabling balanced evaluation of lateral flexion, abduction/adduction, and flexion/extension. Future research should improve estimation performance in cases of body occlusion by expanding datasets and refining algorithms. Additionally, analyzing tilt caused by distance variations and applying a correction ratio could help mitigate MediaPipe’s tilt issue. Identifying optimal camera placements can enhance the accuracy of monocular markerless 3D motion capture, supporting its broader clinical application.

Keywords

• 3D motion capture
• marker-based
• monocular markerless
• range of motion
• mean absolute error
• correlation