摘要
结合柔性上肢康复机器人辅助患者进行上肢康复时的位姿检测需求,文章基于YOLOv8-Pose改进了上肢位姿检测模型。引入FasterNext模块,丰富了所提取的特征信息,提升了检测精度;引入LADH检测头,在保持计算效率的同时,减少了参数量;引入PIoU损失函数,加速了边框检测收敛,减少了漏检率,提高了检测精确度。验证结果表明改进的人体上肢位姿检测模型能够提高位姿检测的精度与速度,可以满足柔性上肢康复机器人的应用控制需求;相较于原型YOLOv8-Pose,改进模型的P值、R值、mAP@0.5、mAP@0.5:0.95依次为84.1%、85.1%、83.2%和43.3%,分别增加了1.2%、2.1%、6.3%和3.8%;GFLOPs下降至8.1,减少了6.9%;参数量下降至2.80,减少了12.6%。研究成果可为人体位姿检测提供一定的参考。
This paper improves the upper limb pose detection model based on YOLOv8-Pose,to address the pose detection needs of flexible upper limb rehabilitation robots to assist patients in upper limb rehabilitation.By introducing the FasterNext module,the extracted feature information is enriched and the target detection accuracy is improved.By introducing the LADH detection head,the number of parameters is reduced while maintaining computational efficiency.By introducing the PIoU loss function,the bounding box detection convergence is accelerated,the missed detection rate is reduced,and the detection accuracy is improved.The model verification results show that the improved human upper limb pose detection model can improve the accuracy and speed of pose detection and meet the application control requirements of flexible upper limb rehabilitation robots.The P value,R value,mAP@0.5,and mAP@0.5:0.95 of the improved model are 84.1%,85.1%,83.2%,and 43.3%,respectively,which are 1.2%,2.1%,6.3%,and 3.8%higher than those of the prototype YOLOv8-Pose.GFLOPs drop to 8.1,a decrease of 6.9%.The number of parameters drop to 2.80,a decrease of 12.6%.The research results can provide a certain reference for human pose detection.
作者
胡晓
周宇
张尧尧
HU Xiao;ZHOU Yu;ZHANG Yaoyao(College of Mechanical Engineering,Chongqing University of Technology,Chongqing 400054,China)
出处
《现代信息科技》
2025年第14期49-54,共6页
Modern Information Technology
基金
重庆市人工智能技术创新重大主题专项重点研发项目(cstc2017rgzn-zdyfx0010)。
