Research On Moving Object Detection And Tracking In Traffic Scene

Smart city system which based on computer vision has important research value and application prospects in several domains which include pedestrian flow analysis of key areas,real-time tracking and the intelligent guidance of traffic flow.And moving object detection and tracking is the key technology to achieve follow-up intelligent analysis and decisions.In view of the accuracy of the background model initialization which has a significant impact for object detection effects,this paper focused on the ghost issue which happens easily and difficult to solve during initialization process.And an improved inter-frame difference method was put forward which would be fused with internal and external contour histogram matching method,and take full advantage of spatio-temporal information of the video sequence.Also taking into account the moving object occlusion would lead to sharp decline in object tracking performance,so a method with occlusion judgment was put forward in this paper which is based on the classifier maximum corresponding decay rate,and trigger conditions on different levels were used.It would significantly improve the efficiency and accuracy of target tracking.Finally,the detection and tracking applications of vehicle and pedestrian were taken for example respectively,and the new methods which were above-mentioned were carried out trial application and the feasibility of the method was verified.The main work of this paper is as follows:(1)On moving target detection,in view of the background model initialization process of ghost problem,put forward the ghost suppression method based on the characteristics of time and space.The method solves the two problems,including the internal and external contours histogram matching method cannot be applied to small ghost area and the inter-frame difference method is sensitive to camera dithering.The traditional inter-frame difference method places undue reliance on the fixed threshold.This paper uses the normal distribution of the inter-frame result to represent the inter-frame characteristics,and the applicability is improved.The experimental results show that the proposed ghost suppression method based on the characteristics of time and space is more effective.Finally in view of the fact that there are some holes and noise in the results,using morphological processing and the operation of removing small area to optimize the results.(2)On moving target tracking,in view of how to determine hide problems,this paper puts forward a a method with occlusion judgment based on the classifier maximum corresponding decay rate.The method which is based on decay and histogram matching degree divide the process into two stages.The experimental results show that,compared with the judgment method based on Surf feature matching and the judgment method based on LK optical flow,the method is more effective.(3)In order to optimize the results of pedestrian tracking,we apply the technology of static pedestrian detection to target tracking.Enlarging the pedestrian tracking window appropriately as area for optimized,with the applying of pedestrian classification detector which has been trained,we affirm the location of target finally.The experimental results showed that the optimized pedestrian tracking results are more accurate.(4)Using the relevant technology of moving target detection and tracking for the design of experiment and analysis.The experimental system tracks the target(vehicle and pedestrian)automatically who enters the area of interest.The results shows the effect of the proposed approach.