Highlight moving object detection based on spatiotemporal saliency in dynamic background

Abstract

Changing area extraction from the sequential images is the main role of motion detect, the detection result will be seriously affected by the dynamic background, so that the efficient motion detection becomes a difficult task. Inspired by the saliency detection, this paper presents a new method for moving object detection, combined with the bottom-up and top-down visual computing model to obtain spatiotemporal saliency. The proposed method detected the spatial saliency map, and adds a time dimension on its basis to obtain the temporal saliency map by using the idea of a three-frame difference to detect the salient moving object in the video. The proposed method is simple and effective to detect salient motion region. Experiment results shown that the detected moving object was salient in dynamic background, and good to overcome the difficulties in special circumstances in which camera shakes ceaselessly. This method has widely applicable and better robustness.