运动捕捉人工智能系统在速滑项目中的应用

为验证自主研发的人体运动自动捕捉人工智能系统在大范围运动场景下获得人体关节点三维坐标的有效性，在训练环境下对10名速度滑冰运动员的动作特征进行三维定点拍摄，使用该系统对录像进行自动解析，获得的三维坐标-时间曲线和专业研究人员人工解析曲线的多重相关系数均大于0.90，差值平均值小于0.025 m，验证了该系统在大范围运动场景下快速、精确获得人体关节点三维坐标的有效性。将该系统应用于我国顶尖速度滑冰运动员的动作技术分析，结果表明成绩差异主要来自蹬冰效果的不同，提示运动员应着重改善蹬冰动作，提高蹬冰效率。

Application of the Artificial Intelligence System for Human Motion Automatic Capture in Speed Skating

In order to validate the self-developed artificial intelligence system for human motion capture in obtaining the three-dimensional (3D) coordinates of human joint points in a large-scale motion scene, 3D kinematic data of 10 speed skaters was collected by 3D fixed-point photography in training environment and then the videos were automatically analyzed by the system. 3D coordinate-time curves had a multiple correlation coefficient of > 0.90 with data manually parsed by professional researchers, with the absolute error < 0.025m, which validate the effectiveness of the system in obtaining the fast and accurate 3D coordinates of human joint points in a large-scale motion scene. Applying the system on the movement technique analysis of top Chinese speed skaters revealed that the performance difference mainly comes from the difference of push-off effectiveness, hence skaters should focus on improving the push-off technique and efficiency.