Crowd Density Estimation Based On Video Image

With the rapid development of the global economy, the word is becoming a whole, and gathering personnel become more frequent. Crowd events become more and more, which cause huge losses of people’s lives and property and make a huge market demand based on video surveillance of mass incidents warning. With the rapid development of video image processing technology, the early warning of abnormal mass incidents based on video images is possible now. Crowd density is the main feature of the abnormal mass incidents. Crowd density estimation based on the video image is the first step towards of the abnormal group event warning, and is the most important step, which relates to image processing, computer vision, pattern recognition, and many other disciplines. Crowd density has become a hot and difficult research in the field. In this paper, the methods of the crowd density estimation based on video image are made a profound study, which are based on the previous work. Two good methods which could adapt complex scene are studied. Details are as follows:(1) The structure of mainstream intelligent video surveillance system is introduced, and the core parts are studied. The different types of detection methods of people objects are described, and crowd target detection method based on Gaussian mixture background model is studied. Three image denoising filters are analyzed: smooth filter, mathematical morphology filter, frequency domain filter. For the projective deformity problems in the crowd images, a plug weight algorithm based on linear perspective correction surveillance video is introduced, and the principle and implementation process are analyzed.(2) For sparse crowd scene, the crowd density estimation method based on pixel statistics is introduced, and the principle and implementation process are analyzed. The crowd target detection method based on Gaussian mixture background model is introduced, and two improvements are proposed: increase spatial information, change the dynamic weight update function.(3) For dense crowd scene, the crowd density estimation method based on texture analysis is introduced. The crowd features extraction method based on GLCM is introduced, and contrast, correlation, energy and homogeneity are used to measure crowd image texture features.(4) The different types of people videoes are collected. These two algorithms are simulated, and a detailed analysis and comparison are made.