Crowd Density Estimation And Abnormal Behavior Detection

The security of public places has always been a hot spot of attention all over the world.In recent years,with the rapid development of China's economy and the increasing population,the security problems of public places are becoming more and more prominent.At present,the monitoring system has been widely used in all public places,for example,the airport,the railway station,etc.With the rapid development of the field of computer vision,intelligent video analysis technology has been widely concerned by scholars.In particular,how to use computers to detect the state of people in a scene is one of the hot topics in the current study.This paper focuses on crowd density estimation and crowd abnormal behavior detection.The innovations and achievements of this paper involve three aspects:(1)To overcome the influence of crowd density and features utilized in crowd counting applications,a novel crowd counting algorithm based on density classification and combination features is proposed.This algorithm includes two phases: offline combination features selection and on-line real-time estimation.In the offline phase,a density threshold is specified to classify the image samples into two categories.The optimal combination features are then determined through experimental methods.In the on-line phase,firstly,all images are classified into high and low density categories using a classifier.Then,using the corresponding Gaussian model trained with the selected features in the offline phase,the two groups of images are estimated,respectively.Compared with the current mainstream estimation algorithm,the average estimation error of the proposed algorithm is better.(2)For abnormal crowd event detection,an algorithm based on the combination of space and time is proposed.From the spatial perspective,the average kinetic energy distribution histogram is extracted to describe the crowd movement characteristics,and support vector machine classifier is used to classify the crowd movement characteristics;From the time perspective,the hidden Markov model is built to detect the continuous crowd behavior in the scene.Experimental results on the UMN benchmarks show that our algorithm can effectively detect crowd anomaly behavior.The detection results of our algorithm are better than traditional optical flow method,SIFT point detection method and social force method.(3)In order to control the accident scene and perform resource scheduling,it is necessary to locate the source of the incident.An event source localization algorithm for abnormal crowd event is proposed.Under the framework of RANSAC(Random sample Consensus,RANSAC),the location of the event source is realized by calculating the intersection point of the reverse extension line of the abnormal behavior crowd movement,where the multiple event sources can be labeled simultaneously,our algorithm successfully localize the event sources.Experimental results on the UMN benchmarks show that our algorithm can successfully localize the event sources.