Research On Video Surveillance Moving Target Analysis Based On Gaussian Background Model

Video based moving object analysis is widely used in road traffic,airports,banks and other places with large traffic flow and high requirements for safety performance,to detect,track and shadow suppression of video image sequences containing various moving objects in the scene,and to understand and describe their behaviors.With the rapid development of computer field,the time cost of target detection has been greatly reduced.With the increasing demand of image processing technology,the updating and realization of fast and efficient moving object detection algorithm is of great practical significance for the development of video understanding technology.So far,researchers have made a lot of achievements in the field of moving target detection.These achievements come from the use of the advantages and disadvantages of the algorithm and the video image environment.But so far there is no one method that can be used in all environmental conditions,which also promotes the exploration of image processing by researchers.In this paper,targets in detection video surveillance are two relatively common target categories: people and vehicles.The main research work includes the following aspects:(1)Video image related knowledge mainly introduces image noise,image filtering and binary image processing.According to the evaluation criteria that the higher the PSNR is and the smaller the mean square error is,the better the denoising effect is,the mean filter,median filter and wavelet threshold transform are respectively used to detect the salt-and-pepper noise and Gaussian noise in the video.(2)Moving target detection in view of the traditional,block,add the life value of the single gaussian and traditional,the improved adaptive learning rate of background subtraction division the gaussian mixture model respectively for people and vehicle video frames for target detection and denoising processing,through statistical histogram pixel,the algorithm running time,target detection effect of different gauss model to analyze different goals.(3)Shadow suppression adopts HSV color space,which is close to human color perception,and converts RGB color space into HSV color space to remove the shadow region in the acquired target.HSV and texture features and HSV and gradient features were used to suppress the shadow of people and vehicles,respectively.The detection and evaluation of shadow detection rate,discrimination rate and the average value of the sum of the two were carried out.(4)Two commonly used tracking algorithms are adopted for moving target tracking: Mean Shift tracking algorithm and Cam Shift tracking algorithm,while the Cam Shift tracking algorithm adding saturation component and brightness component of the Cam Shift tracking algorithm to compare the target tracking,observe the moving target tracking trajectory coherence and target tracking recognition accuracy to evaluate the effectiveness of the tracking algorithm.The experimental results show that,compared with the traditional algorithm,the improved single Gaussian background model and the improved mixed Gaussian background model target detection algorithm have different detection effects on different targets.Compared with mean filtering,median filtering,soft threshold and hard threshold,wavelet semi-soft threshold combined with mathematical morphology closure operation has the best processing effect on the two kinds of noise.HSV and texture features and gradient features have different effects on shadow suppression for different targets;Cam Shift tracking algorithm with added saturation component can effectively improve the problem of missing tracking caused by target occlusion.