基于集群性特征的异常行为检测

Abnormal Behavior Detection Based on Collectiveness Feature

在异常行为检测中,群体行为难以描述。针对该情况,提出了一种基于个体与群体中其他个体的行为相似性(集群性特征)的异常行为检测方法。该方法首先采用混合高斯模型提取出视频的背景;然后,使用KLT(Kanade–Lucas–Tomasi)算法追踪运动人群;接着,从群体的运动方向和速度两个角度提取出集群性特征;最后,利用集群性特征直方图描述行为,计算直方图的熵值来判断行为的异常。基于不同场景下的视频序列所进行的测试结果验证了所提方法的有效性。

Among the abnormal behavior detection methods, it is difficult to describe the crowd behavior. For this case, an abnormal behavior detection approach based on behavioral similarity （collectiveness features） between individual and other individuals in the group is proposed. Firstly, Gaussians mixture model was used to extract the background of the video. Then, Kanade-Lucas-Tomasi （KLT） algorithm was used to track the moving crowd. Next, collectiveness features integrated the motion information of the whole crowd are extracted from the direction and speed of the crowd movement. Finally, a histogram derived from the collectiveness features was defined to measure the anomaly of crowd activity, and the entropy of the histogram was computed to recognize abnormal events. Experiments were conducted on various video datasets, and results were presented to verify the effectiveness of the proposed scheme