公共场所人群加速度异常检测系统

Renovated abnormal passenger crowd behavior detection system based on the speeding-up and squeezing in the public places

针对目前基于速度检测公共场所密集人群异常行为存在的检测准确率低、使用范围局限的问题,从人群的加速度角度对可能导致公共安全事故的人群异常行为进行研究,提出了一种基于加速度检测人群异常行为的算法,并基于该算法实现了针对人群逃散、人群聚集、人群拥挤和人群逆行4种异常行为检测的系统。首先,利用金字塔Lucas-Kanade光流法进行特征点跟踪;然后,在获取到特征点的速度矩阵基础上进一步计算其加速度矩阵,反映速度的整体变化;最后,从加速度大小和方向两方面检测人群异常行为。结果表明,所提算法检测用时较少,相比基于速度检测的对比算法,检测的正确率提高到80%,误报率降低为5%。

This paper is aimed to improve the detection accuracy of the walking speed-based abnormal crowd behaviors so as to broaden its application scope. For this purpose,the author has made an investigative statistics on the irregular crowd behaviors,which may lead to the public safety accidents due to the upspeeding. As a result of our study,it has been found that such irregular crowds may turn to be in a mess in case of emergency comes about,for most of the passengers turn to speed up at a rate beyond the regular changing speed,with their moving directions tending to be chaotic and instable. Therefore,in accordance with the characteristic features we have gained,we have developed a new algorithm for inspecting such abnormal crowd behaviors by way of the speeding-up features in response to such changes both in speed and direction. The aforementioned algorithm can help to track the feature points first of all with the pyramid Lucas-Kanade optical flow method,and,then,has eabled us to work out the speeding-up matrix of the tracked points to get the overall changes of the velocity,and eventually detect the irregular behaviors by examining the magnitude and direction of the acceleration. Thus,the proposed algorithm can help to capture the irregular crowd behaviors quickly and meet the need of real-time detection. As compared with the current velocity-based detection algorithm,the proposed algorithm is expected to increase an evaluation accuracy rate of 80% as well as to reduce the false positive rate by 5%.Therefore,it can be concluded that the proposed algorithm we have developed can help to improve an accuracy significantly and better interpret the actual migration of the passengers. In addition,the paper can also be able to execute a new detecting system with the algorithm we have proposed to detect the four types of the irregular crowd behaviors,so as to offer a great potential for the application to the public safety secure systems.