八段锦运动场景下的手部关键点检测算法研究OA

Objective To design a hand key point detection algorithm for the intelligent guidance and evaluation system of Baduanjin,in order to solve a series of challenging problems such as occlusion,complex action details and mobile deployment.Methods Based on the YOLOv8n-pose model,a hand key point detection algorithm YOLO-BDJ-Hands for the Baduanjin scene was proposed.The separation enhanced attention module was introduced to learn the adaptive weight attention of hand occlusion region.The receptive field expansion module was used to increase the receptive field range and enhance the ability to detect hand detail features at different scales.Results On the Baduanjin hand key point dataset,the detection precision and recall rate of the YOLO-BDJ-Hands algorithm were 0.912 and 0.901 respectively.The number of model params was 3.1 M,which was 9%less than that of the original model.The floating point operations was 9.5 B,which could meet the requirements of mobile deployment.Conclusion The YOLO-BDJ-Hands algorithm proposed in this study has high robustness and generalization performance,basically achieving model lightweighting,facilitating deployment and application on mobile terminals,and can provide effective technical support for the subsequent development and promotion of the intelligent guidance and evaluation system of Baduanjin.