摘要：

现有的人群异常检测算法多数存在未对异常事件进行评估、无法对异常点定位以及实时性差等问题。为此,提出一个基于欧拉法的人群异常检测算法。通过叠加连续的差分图获取累积差图像(ADI),将ADI的梯度场代替速度场,可使ADI转换为二值图像以减少外部干扰,并将ADI梯度场与二值图像的点乘和作为人群能量值。通过分析人群能量的变化,可实时地对人群异常事件进行报警、分级评估和定位。在PETS2009数据库上的实验结果表明,该算法的识别率超过97%,平均每帧的处理时间为0.01 s,分级评估正确,异常点定位与人工定位基本重合。

关键词: 智能视频监控, 人群异常检测, 流体力学, 欧拉法, 事件分级, 异常点定位

Abstract:

Most existing crowd anomaly detection algorithms have problems such as no assessment of abnormal events,inability to locate abnormal points,and poor real-time performance.An algorithm based on Eulerian method for crowd anomaly detection is proposed.Accumulative Difference Image(ADI) image is obtained through overlying continuous difference image,the gradient field of ADI is substituted for the velocity field,so that ADI is converted into a binary image to reduce the external interference,and the point multiplication of ADI gradient field and binary image as population energy value.By analyzing the changes of the population’s energy,the algorithm can alarm,assess and locate the crowd abnormal events in real time.Experimental results on the PETS2009 database show that the recognition rate of the algorithm is more than 97%,the average processing time of per frame is 0.01 s,the grade assessment is correct,and the abnormal point positioning and manual positioning are basically coincident.

Key words: intelligent video surveillance, crowd abnormal detection, fluid mechanics, Eulerian method, event classification grade, abnormal points location

中图分类号: 

• TP391