Moving Objects Detection Based on Background Updating and Spatiotemporal Saliency Map

Abstract

In this paper, we study moving objects detection in complicated traffic crossroad scenes and put forward a new approach which combines background updating and spatiotemporal saliency detection. When dealing with complex scenes, the learning rate of background updating can not be set too high, which leads to delayed background updates. On the other hand, current spatiotemporal saliency detection can not only detect the moving objects, but also detect the static and odd objects. To solve this problem, we firstly propose a new set of spatiotemporal saliency feature to eliminate the influence of static and odd objects, and then combine the spatiotemporal saliency map and Gaussian mixed model foreground to obtain the final foreground mask. The experimental results show that this approach has better performance than Gaussian mixed model and spatiotemporal saliency detection in moving objects detection.