中文摘要：

提出一种基于运动轨迹矢量化的实时出拳动作分割和拳法类型识别方法，运动数据由一个7相机的Eagle运动捕捉系统提供.该方法实时分割出连续运动序列中的单个出拳动作，并对持续时间可变的出拳动作进行统一矢量化表示.统一矢量化后的出拳动作样本被用来训练一个基于LDA的拳法类型识别器对拳法类型进行实时在线识别.本文实现了一个实时在线拳法类型识别的原型系统，拳法类型识别的正确率达到98%.

英文摘要：

This paper proposes a method for real-time segmentation and recognition of typical types of punches from continuous captured boxing sequences. Real-time motion data is real-timely provided by a 7-camera Eagle motion capture system. Then segmented punches of variable length are vectorized and uniformly represented. These uniformly represented punches are used to train a LDA based punch type recognizer. Recognition accuracy of the proposed method is 98%. A prototype system for realtime online punch type recognition is implemented in this paper.