Target Motion Analysis Of Complex Scenes Based On Video Surveillance

Intelligent video surveillance technology is one of the hot spots in the current research,which is related to the maintenance of social stability,production safety,people's harmonious life.Target detection and motion analysis is a hot research topic in the field of intelligent video surveillance.In this paper,the target detection and motion analysis are studied.Target detection is the basis of video surveillance,which is related to the following research.Firstly,this paper introduces the technology of moving object detection in video processing,and analyzes their advantages and disadvantages.An improved method is proposed to solve the problem of illumination mutation in the mixed Gauss model which is combined with three frame difference algorithm.The effect is much better than the original Gauss mixture model.Then,a kind of background updating algorithm based on sliding window for Gauss mixture model is proposed,which has no obvious adverse effects on not only background modeling or the detection of foreground,but also accelerate the efficiency of the algorithm to provide more time resources.Then,this paper presents a method for calculating the optical flow of the target object based on the moving region.Secondly,this paper introduces two kinds of feature descriptors in computer vision:HOG features and Haar-like features from the point of view of feature information detection.The feature training classifier is extracted from the samples of the specific target,which can detect the target object in the surveillance video.In the research of vehicle motion analysis in traffic scenes,this paper makes use of the characteristic information(model and color)to select the vehicle for specific models and colors and to statistical the traffic flow.In order to reduce the influence of complex environment and multi object,this paper combines motion information and feature information to detect vehicle.The experimental results show that the combination of motion information and feature information can not only eliminate the influence of complex background,but also improve the recognition speed.In the study of human motion in indoor scenes,the problem of occlusion is solved by taking two cameras as an example.In the experiment,the location of pedestrians in the scene are detected by combining the motion information and the characteristic information.Finally,from the point of view of the characteristics of the optical flow,four kinds of motion characteristics are extracted to train classifier to distinguish different behavior.The experimental results show that the proposed method has better recognition results.Finally,the work of this paper is summarized,and the deficiencies in this paper and the future work are put forward.