Design And Research Of System Used For Monitoring Crowd Density

In crowded areas, excessive concentration of the crowd has brought a huge potential safety hazard, so it is an urgent problem to monitor the crowd density in these areas which has great guiding significance for relevant departments to dispatch public traffic, ensure public safety and disperse the crowd as soon as possible.In this paper, the crowd density estimation algorithm based on pixel-counting and texture feature are studied through monitoring actual videos and the method of pressure detection is designed as a supplement program for video surveillance. The main work of the paper can be summarized as follow:(1) Firstly, each step of the crowd density estimation algorithm based on pixel-counting is studied and various methods of moving targets detection, noise elimination, background modeling and edge detection are discussed in detail. Then, this algorithm is used to estimate the crowd density of actual shooting video and analyzing the final result. The conclusion is drawn that this algorithm has a high accuracy in monitoring the low density crowd.(2) The target image texture features are extracted by constructing a gray level co-occurrence matrix and the influences on texture feature values made by different pixel distances, directions and gray levels are analyzed. Then using SVM method to train samples-images of 5 crowd density levels and using classification models to predict crowd density levels of test images. According to the accuracy of prediction, the crowd density estimation algorithm based on texture feature has a high ability to identify the high density crowd. Finally, the reasons for some test sample categories which are forecast by mistake are analyzed.(3) According to the disadvantage of video surveillance susceptible to the environment, the pressure detection method is proposed to supplement the lack of video surveillance in some special occasions. Through extracting pedestrian footprint feature in pressure testing equipment, using the noise elimination method, footprints segmentation algorithm and pedestrian directions discrimination algorithm to count pedestrians of different directions, then using the software to simulate the actual situation, the simulation result shows that this method can accurately count pedestrians in different directions.In conclusion, in this paper, three methods are used to monitor the crowd density and the prospective tentative is achieved which has a certain reference value for the actual application.