基于组合特征的人体动作识别算法研究

Action recognition based on the combined feature

提出了一种人体重心变化特征与改进的Zernike矩特征组合的方法,应用于双层隐马尔可夫模型中进行动作识别。重心变化特征描述了运动目标在空间中的轨迹,反应了人体动作的概略特征;改进的Zernike矩特征有计算速度快、克服敏感区域等特点,刻画了肢体相对位置等细节特征,它们的组合能充分提取视频中运动目标的信息。在模型选择中,选用双层隐马尔可夫模型,该模型可以表达特征之间的内在关联。采用Weizmann视频数据库进行实验,结果证明文中提出的算法具有较好的识别效果。

A kind of feature that combined the human body＇s trajectory of barycentre and the improved human posture of Zernike moments was proposed in this paper and then was applied into the double hierarchy of HMM. The trajectory of barycentre depicts the moving path of human body in space, and generally reflects the characteristics of the action. The improved human posture of Zernike moments can be computed fast and overcome the sensitive area. At the same time it can describe the physical details such as the relative positions. The combination of the two features can extract the information of human action in the image sequence completely. In the experiment, the double hierarchy of HMM was applied which can express the internal connection between two features to classify the action. By testing on the Weizmann database, the result proved that the proposed method could realize the higher recognition accuracy.