Pedestrian Counting And Anomaly Detection Of Crowd Based On Video

As the continuous development of economy, people’s safety consciousness is improving,and the increasing severe domestic and international anti-terrorism situation, people are ofhigher requirements for the safety in public places. So far, there have been several domesticviolent terrorist attacks, which has caused a large number of casualties and property losses,establish efficient complete intelligent video surveillance system has become the urgent needsin today’s society. While the video monitoring system, which is widely used at the presentstage, provide a large amount of video data information, but it is not capable to pre/alarm foremergencies, human must participate in the monitoring work. As the machine visiontechnology and image processing technology progress, the traditional video monitoring systemthat need a lot of manpower already cannot satisfy the needs of social development, a newgeneration of highly automated and intelligent video monitoring system will gradually replacethe traditional video monitoring system in the field of security, while guarantee theperformance of the system, human liberation, thus reduce the cost.In the view of pedestrian safety in public places, to avoid the casualties caused byemergencies, we put forward a people counting and abnormal behavior detection methodbased on video, don’t need to a single pedestrian segmentation as well as the training sample.In terms of pedestrian detection, we use background subtraction method, and VIBE algorithmis used to background modeling, and through a series of means, such as binaryzation,morphology processing, realized with moving pedestrian detection and extraction in the scene.Based on motion foreground extraction, on the basis of a single person’s energy, the number ofpedestrians in monitoring scene can be obtained more accurate. At the same time, we usedistributive entropy and average speed of crowd movement as overall features. On the onehand, through the background modeling of the monitor scene, the foreground is exacted.According to the spatial distribution of the foreground pixels, the distribution entropy isdefined and used to measure the concentration of the crowd and implement detecting thecrowd gathered. On the other hand, the motion vector of the feature points between twoconsecutive frames in the video sequence is gained by tracking the feature points on the image using the Optical flow method, which is used to estimate the crowd’s speed and detect thecrowd running. Experimental results verify the effectiveness of the proposed method.