基于HMM的羽球动作实时识别

Real Time Recognition of Badminton Action Based on Hidden

机器学习与人工智能的快速发展,在人体动作分析与识别领域发挥着日益显著的作用。论文采用粘贴在羽毛球拍柄的单个加速度传感器进行羽球动作的数据采集,使用滑动窗口进行击球信号提取,提出了动作分帧结合K-means等无监督式学习算法进行聚类分析和矢量量化。通过建立隐马尔科夫模型(HMM),改进训练算法对羽球动作进行高效识别。实验表明,论文所设计的系统对8种常见的击球动作进行实时识别,识别率可达94%。

Machine learning and artificial intelligence have made rapid progress recently. They also play a significant role in the field of human motion analysis and recognition. In this paper,a single acceleration sensor is attached to the end of a badminton racket handle to collect its movement data. A sliding window technique is used to extract each hitting action,which is further divid. ed into several sub-action. Combining unsupervised learning algorithm such as K-means for the clustering analysis and vector quan. tization with a hidden Markov model(HMM),the improved training algorithm is able to recognize badminton actions efficiently. It shows a real time recognition rate of 94% for 8 common actions.