Study On Background Estimation And Updating Algorithm Used In Video Traffic Analysis

Background subtraction method is usually adopted to detect vehicle target in video traffic detection system. And the key of realizing background subtraction method is the acquisition of initial background image and dynamic update of background image.This paper introduces some common background estimation algorithms, and analysis their effectiveness and shortcomings with video datas collected from scene, and then presents a new background estimation algorithm. This algorithm is a combination of statistic histogram method and statistic mid-value method. It firstly finds the mid-value after eliminating the abnormal gray in long-range view, and then finds out gray value which has the maximum frequeny around the mid-value as the current pixel gray of the background. This algorithm not only retains the advantages of statistic method such as simple model, anti-noise, and anti-jamming, but also greatly improves the disadvantage of statictis algorithm which is easy to wrongly estimate the gray value of the light and the body of the car as the background gray.In practical application, due to changes in lighting conditions and scenes, the background is dynamic. If wants to ensure the system can work normally in such changes, the background must be updated in real-time. In this paper, two background updating algorithms are presented for tunnel-type scene and the road-type scene. Among them, the algorithm used for the tunnel is based on the initial background, and according to the SAD value of corresponding block between adjacent frames to determine whether access to renewable state, and ues of binarization results with adjacent stable states to determine whether timely update or delayed update. The algorithm used for the road divids the background into effective background and invalid background, and uses the SAD of corresponding block between the current frame and template image to judge whether turn into stable state, and connects adjacent stable states which have small differences, and then determine whether can turn into renewable state.Test the background estimation algorithm and background updating, algorithm for tunnel-type scene presented in this paper with video datas collected from Shanghai Fuxing Road, the results show that:the background estimation algorithm can commemdably estimate the background when traffic condition is smooth, and the background updating algorithm not only can timely update the changes of light, but also can achieve delayed update when targets stays to meet the system demand for follow-up treatment. Test the background updating algorithm for road-type scene with video datas collected from Shanghai Outer Ring, the result shows that:even in the case of large traffic flow, this algorithm also has satisfactory results. The algorithms presented in this paper have been appied successfully in actual traffic detection system, and preliminary application shows that they can adapt to more complex traffic environment.