An Enhanced Background Estimation Algorithm for Vehicle Detection in Urban Traffic Video

This paper proposes an enhanced version of the sigma delta background estimation method, suitable for urban traffic scenes. In the original algorithm, the background model quickly degrades in such complex scenes, being easily contaminated by slow moving or temporarily stopped vehicles. Some heuristics have been added to the basic algorithm in order to make a selective background model updating at the pixel level. Experimental tests made over typical urban traffic streams prove the validity of the proposed enhanced version.