Video Surveillance Technology Based On Population Density Estimation

Along with acceleration of the process of urbanization in China, the scale of Chinese cities expands constantly, which leads to the increasing risk of massive tramp accident in large assembly, transportation hub and other crowd place. We can take population density as a indicator for population environment risk. There're too many shortcomings for traditional manual monitoring such as the strong subjectivity, low reliability, and short effective surveillance time which decided by physiological characteristics of human. Intelligent video surveillance technology can monitor the crowd in real time, and classify the levels of population density, and make early-warning for the risk of possible accident.This paper makes a further research on the core algorithm of intelligent video surveillance technology which is applied to the crowd density monitoring. Optimization algorithms based on the existing two mainstream population density estimation algorithm is proposed. One of them is based on pixel statistics.This paper use multiple linear fitting and BP neural network to describe the linear and nonlinear relationship between the pixel eigenvalues and the population size, and compared the effect of them in different population density environment. For the second types of crowd density estimation algorithm based on texture analysis, In this paper, we classify the high population density by taking advantage of the characteristic of GLCM and the classification function of SVM. At the same time, taking into account the high dimension of the texture extracted from the images, which leads to the increase in the amount of data processing, and is not conducive to real-time video processing, In this paper, principal component analysis(PCA) is used to exact the one-dimemsion texture statistics, three- dimemsion texture statistics and twelvedimemsion texture statistics, and compare the accuracy of them.