Dynamic Background Reconstruction in Traffic Surveillance Systems

Abstract

In the Vision-based systems for traffic surveillance, background suppression is usually used for vehicle detection. Therefore the accuracy of detection is heavily based on the exact acquirement of background. This work presents a technique for dynamic background reconstruction. The course of it is shown as follow: first of all, Pixel of background is detected by using T distribution which is used to eliminate the effect of noise. During the processing the variance of camera is not needed. Then we reconstruct a frame of background which doesn't contain any moving objects by integrating of several frames of image. And when we detect vehicle in the latter, we update the background at the same time, resulting from changing of background gradually with time going. The experimental results show that the background reconstructed reflects the real background.