基于粒子熵值的异常行为检测

Abnormal Behavior Detection Based on Particle Entropy

在异常行为检测当中,人群的分布信息十分重要。针对该情况,提出了一种基于粒子熵值的异常行为检测方法。该方法采用混合高斯模型动态建模提取出视频图像的背景,在提取到的背景图像上使用KLT(Kanade-LucasTomasi)算法追踪前景获得人群的速度和位置。基于人群粒子的网格分布获取相对应的直方图,并通过计算直方图的粒子熵值描述人群行为;最后,结合粒子分布的熵值与人群粒子的速度,提高异常行为判断的准确性。基于不同场景下的视频序列所进行的实验测试结果验证了所提方法的有效性。

In the abnormal behavior detection,the distribution information of the crowd is very important.Aiming at the situation,an abnormal behavior detection approach based on particle entropy is proposed.The approach uses the gaussian mixture model to extract the background of the video image, KLT （Kanade-Lucas-Tomasi） algorithm is used to track foreground and to obtain the speed and location of crowd;The histogram of the crowd particle distribution is obtained based on the spaee grid where the particle locates. The crowd behavior is described by calculating the histogram particle entropy; finally, by combining the entropy of particle distribution and the speed of crowd particle,tbe algorism improves the accuracy of abnormal behavior detection.The experiments are conducted on various video datasets,and the results are presented to verify the effectiveness of the proposed scheme.