Coarse-to-fine sample-based background subtraction for moving object detection

Abstract

Background subtraction is crucial for detecting moving objects in videos. However, most background subtraction methods usually fail to deal well with illumination changes, static foreground, ghosts, and dynamic background. In this paper, we propose an efficient coarse-to-fine sample-based algorithm to address these issues simultaneously. Specifically, a rough motion region estimation method, is first proposed by combining an improved frame difference method, the block-based image partition algorithm and multi-scale region-based technique to locate the possible area of moving objects. It can compensate the global illumination change and suppress dynamic background. Then, an improved Vibe algorithm with adaptive distance threshold and time sub-sampling factor parameters for each pixel is proposed to detect precise moving objects. It can not only retain static foreground for a long time while eliminating ghosts rapidly, but also detect camouflaged foreground efficiently. Experimental results demonstrate that our proposed algorithm can extract moving objects more accurately with less time cost than several state-of-the-art background subtraction approaches.