基于姿态分解的动作综合评价研究

Posture Segmentation Based Comprehensive Assessment of Actions

基于运动数据的动作定量评价是体育科学智能化发展的基础,但是传统的动作评价方法通常是将待评价的动作信号或模式与标准样式进行整体匹配和分析,很难抓住动作的本质特性,也难以对动作进行精细化评价和分析。为了从结构和本质特性上对动作进行综合评价,本文首先采用SWAB(Sliding Windows and Bottom-up)曲线切割算法对动作过程进行细粒度的分割,使之分解成一列姿态片段,然后针对各姿态片段从标准性、速度和完整性方面进行综合评价,并最终得到动作的整体评价指标。由于姿态片段对比中存在着关键点匹配问题,本文将其转化成为一个动态规划问题并通过传统的优化算法进行求解。在基于传感器的羽毛球动作数据集上的实验结果表明,本文所提出的基于姿态分解的动作评价方法能够对动作进行细粒度的综合评价,并且能够有效地指导运动员在日常练习中提高动作的准确性。

Sport data based quantitative assessment of actions is the basis of the development of intelligent sport science.However,conventional action quantitative assessment methods usually match the pattern or signal of an action with its standard version,and therefore cannot grasp the essential attributes and cannot assess the action in a refined and comprehensive way. In order to assess an action from its structure and essential attributes,we firstly implement the SWAB(Sliding Windows and Bottom-up)curve segmentation algorithm to perform a fine-grained segmentation on the process of the action such that the action is decomposed into a number of posture fragments. We then assess each posture fragment from the viewpoints of standard degree,speed,and integrity,and finally obtain the comprehensive assessment score of the action. To deal with the difficulty in matching the key points of two posture fragments in the above assessment,we transform this point-matching problem into a dynamic programming problem and obtain the optimal matching manner by solving it with the conventional optimization algorithm. It is demonstrated by the experimental results on a sensor based badminton action dataset that our proposed comprehensive action assessment method with posture segmentation provides more fine-grained comprehensive assessment of sport actions,and can effectively guide players to improve their skills.