一种基于光流场的群体运动特征描述子

A crowd behavior feature descriptor based on optical flow field

为了实现公共区域中的群体异常行为的自动识别,提出一种基于光流场向量的人群运动特征描述子。通过从视频中提出的光流场信息,用位置、运动向量、运动方向和运动强度4个维度构建描述子,使用K聚类算法对描述子进行聚类处理,并对K聚类之后的结果进行了运动角度与强度数据的合并处理,最终获得能够描述场景中人群的全局运动特征。在试验中,使用提出的描述子进行了异常行为检测,并与类似的异常检测算法进行了比较,获得了较好的异常检测效果。

On the purpose of achieving the automatic recognition of crowd behavior abnormality in public field,a crowd behavior feature descriptor based on optical flow field is proposed.This descriptor is modeled from the extracted optical flow information with four dimensional data including coordination,motion vectors,orientation and magnitude,and clustered with K-mean cluster algorithm,then converged by orientation and magnitude,to obtain the feature which can describe the global motion feature of the crowd.In the following experiments the proposed descriptor successfully described the motion feature in video,finally an experiment is conducted by comparing the detection result with other algorithms,which proves the proposed algorithms abtains better anonaly detection performance.