Abstract: A method of head behavior recognition based on the analysis of trajectory is presented. It detectes and trackes the face automatically, and calculates parameters of the trajectory of the face. It can achieve the behavior recognition of the head by parameters’ data statistics and analysis. For the problem of the exchange of the long axis and the short axis of the tracking window at the case of up head and down head, it proposes an improved algorithm of the angle between the major axis of the tracking box and the vertical axis. Experimental result shows that the problem of the exchange of the long axis and the short axis have been successfully resolved, and the algorithm can recognize the behavior of the head correctly.

Key words: tracking box, tracking window, centroid, major axis, behavior recognition, trajectory

摘要： 提出一种基于人脸运动轨迹分析的头部行为识别方法。对人脸进行检测和全自动跟踪，通过计算人脸运动轨迹参数并进行参数数据统计和分析，实现人体头部行为识别。针对抬头、低头时出现的跟踪框长短轴交换问题，通过引入基于跟踪框长短轴比值的约束条件，对椭圆跟踪盒主轴与纵轴夹角的计算方法进行改进。实验结果证明，该方法的正确识别率较高。

关键词: 跟踪盒, 跟踪框, 质心, 主轴, 行为识别, 运动轨迹

CLC Number: 

• TP391