This research aims to evaluate how camera position affects the effectiveness of a 2D markerless motion capture system for shoulder movement assessment.
Training of deep neural network ResNet-50 and evaluation using DeepLabCut for shoulder kinematics analysis.
Testing multiple camera positions to determine influence on motion capture accuracy.
Significant variations in capture accuracy based on camera position observed (p<0.05).
Optimal camera position identified, enhancing robustness of shoulder kinematics assessment.
Training with ResNet-50 yielded improved prediction accuracy compared to previous methods.
Abstract
Deep neural network training and evaluationThe network (ResNet-50, DeepLabCut) was trained to identify the location