A Method of Effective Background Estimation for Video Sequences with Dense Moving Objects

A method is proposed to estimate the background in video sequences with dense moving objects. The velocity field is estimated by optical flow to determine the image areas occupied by moving objects. The background is estimated by an efficient averaging process within the regions excluding the moving objects, which overcomes the foreground-occluding problem in common averaging method in a dynamic environment such as heavy traffic. The experimental results on traffic surveillance videos prove the effectiveness of the proposed method.