摘要
针对深度相机与多目相机在使用环境的限制且穿戴式设备影响运动分析准确性的问题,本文提出了一种一般场景下采用单目RGB相机的无标记点式人体姿态运动分析方法,设备简单,适用性广。姿态估计时,识别出的人体姿态是一个二维平面姿态,根据图像中各个子关节到父关节的向量长度比例与角度变化,将图像中关键点的二维坐标逆推出三维坐标。由人体解剖学相关知识将关键点组合在一起,实现人体骨架化,完成实时运动分析。实验结果表明,由三维坐标变化可以判断关节角度变化,该方法可以实现人体的运动分析。
The depth camera and multi-camera have limited use environment and wearable devices affect the accuracy of motion analysis.In this paper,a markless human posture motion analysis method using monocular RGB camera in general scenes is proposed,which has simple equipment and wide applicability.When estimating the posture,the identified human posture is a two-dimensional plane posture.According to the vector length ratio and angle change of each child joint and parent joint in the image,the three-dimensional coordinates of key-points are inversely deduced from the two-dimensional coordinates in the image.The key-points are combined by the relevant knowledge of human anatomy to realize the skeleton of human body and complete real-time motion analysis.The experimental results show that the change of joint angle can be judged by the change of three-dimensional coordinates,and this method can realize the motion analysis of human body.
作者
林逸非
范丽婷
张阳
LIN Yi-Fei;FAN Li-Ting;ZHANG Yang(School of Mechanical Engineering,Shenyang Jianzhu University,Shenyang Liaoning110168,China)
出处
《机电产品开发与创新》
2022年第4期120-123,共4页
Development & Innovation of Machinery & Electrical Products
基金
教育部产学合作协同育人项目(201902261019)。
