Scoring Method for Table Tennis Gesture Motion Based on Deep Learning
胡建华 1郭建丁 2吴伟美 1曾文英 1刘晓林1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 广东科学技术职业学院,计算机工程技术学院,广东,珠海 519090
2. 西南民族大学,计算机科学与工程学院,四川,成都 610041
折叠
摘要
为了方便乒乓球运动训练和相关体育赛事评分,提出一种乒乓球姿态动作评分方法.通过深度学习神经网络提取骨骼关节点数据,得到一组动作序列的空间角度变化值;通过改进的动态时间规整DTW(dynamic time warping)算法把两组动作序列的关节点数据进行匹配,得到序列之间的距离;根据Mean Shift算法聚类数决定各个骨骼关节点权重系数,计算得到两个动作序列距离的大小评分值.实验表明,该方法能很好地实现人体动作评分,提高了关键关节点权重分值的影响,提高了乒乓动作序列评分的准确度,更加符合专家主观评分标准.
Abstract
In order to facilitate the scoring of table tennis training and related sports events,this paper proposes a scoring meth-od for table tennis gesture motion.The method extracts the data of bone joints through deep learning neural network and ob-tains the spatial angle change value of a set of action sequences.Through the improved DTW(dynamic time warping)algo-rithm,the joint data of the two action sequences are matched,and the distance between the sequences is obtained.According to the Mean Shift algorithm clustering number,the weight coefficient of each bone joint is determined,and the magnitude score of the distance between the two action sequences is calculated.Experiments show that this method can achieve human motion sco-ring well,improve the influence of key joint weight scores,improve the accuracy of table tennis motion sequence scoring,and is more in line with the subjective scoring standards of experts.
关键词
深度学习/人体姿态估计/动态时间规整/动作评分
Key words
deep learning/body pose estimation/dynamic time warping/motion scoring
维普
万方数据