The Algorithm Of Crowd Density Detection In Intelligent Monitoring System

In recent years, with the rapid development of social economy, the urbanizationprogress in new construction system has speeded up, and urban population is higher andhigher. Similarly, the communication and exchange of people ranges more widely. Alarge crowd gathered in public places is common things, which increases the probabilityof crowded incidence. In addition some illegal demonstrations and illegal rallies have aserious influence on the society stability, so the crowd density detection based onintelligent monitoring has been widely studied.This thesis works on the methods and processes of population density detectionbased on intelligent monitoring. First we introduce the research background, researchstatus and research purpose, then analyse the methods extracted the prospect of imagesand proposed the new method that combination of background subtraction and threeframe difference method, finally carry on the statistics of edge pixel. Special emphasisis prospect extraction, we try to contrast a variety of prospect extraction andcomparative analysis.The analysis shows that the pixel-based crowd estimation is simple but theestimated error increases as the population density increases，and the texture-basedcrowd estimation performs better in high population but is morecomplex．Comprehensive analysis of above conclusion, this thesis uses the new methodthat combination of pixel-based crowd estimation and texture-based crowd estimation.First we use pixel-based crowd estimation roughly to divide crowd into low-densitypopulation and high-density crowd, then use the same way to divide low densitypopulation into the low density and lower density. For high-density crowd we usetexture-based crowd estimation, and propose SVM to divided high density populationinto dense, more dense and Congestion. At last, this thesis carries out a series ofexperiments to verify the feasibility of above methods.