The Research Of Th Video Crowd Density Estimation Technology In A Large Square

With the development of society, various social activities (such as meetings,exhibitions, lantern activities, etc.) are increasing frequently, and how to ensuresafety of the people in the place has become a serious problem. This thesis is basedon the facts, it combines the Nanjing Public Security Bureau's research projects,and it is used for crowd density estimation in the central area of Confucius Temple,as a part of security command platform in Qinhuai Branch's International LanternFestival in China ? Qinhuai , through real-time video analysis technology toprovide real-time crowd density estimation for the command platform, when thedensity is more than the setted threshold, the alarm signal is given promptly to thesecurity departments to take relevant measures. Population density is an importantindicator of population surveillance, crowd management is one of the mostimportant basis. Polus [1] have suggested that population density and communitysafety, and people are closely related to the level of service.This paper describes the development of crowd density estimation indomestic and international situation . Through the study we may find that thedensity of pixel-based estimation method is simple, computation, but when thecrowd density is high, the crowd blocking makes the error greater; the using oftexture analysis methods can make full use of the image texture information toconduct density estimation . This method can solve the problem of high-densitypopulation density,but when the background is complex, the density estimationerror is very large, what's more, the high computational complexity is not conducive to real-time video surveillance. In this case, the paper presents a fuzzymatching method based on adaptive density estimation algorithm, which accordingto polus [1] who proposed the theory that the density is divided into five levels,namely sparse, normal, saturated, crowde and alarm. The test proved that themethod can get more accurate classification results, get faster running speed, andprovides a good density estimation method for real-time video surveillance .