Posture angle test based on RGB-D neural network for ice and snow sports elbow pads
Lijun Wang, Jiyang Mao, Xianzhong Chen, Moyang Chen, Jiale Deng · 2022
A target recognition method based on RGB-D network and calculated the posture angle by measuring the markers on the prosthesis arm to address the problem of real-time measurement of prosthetic posture angle for the analysis of the motion effectiveness of existing ice and snow sports elbow pads is proposed. Firstly, on the basis of YOLOv5 network, a dual-channel RGB-D network model is designed to recognize the posture of wearing the elbow pads prosthesis arm and analyze the motion effectiveness of the ice and snow sports elbow pads. Secondly, recognition experiments are conducted under different lighting conditions, and compared with YOLO V5, Faster RCNN target detection model, the designed network recognized 8.5% and 7.2% improvement in mPA, precision is improved by 5.9% and 3.3% respectively. Finally, after applied to the pose angle analysis of the ice and snow sports elbow pads, it is proved that the dual-channel RGB-D network can effectively improve the accuracy, speed and robustness of target recognition, and can be used for real-time target detection. The designed network model can be applied to the posture angle test in the ice and snow sports elbow pads movement effectiveness test to assist in analyzing the ice and snow sports elbow pads movement effectiveness.