Background Generation Using Mean-Shift and Fast Marching Method

Background generation is very important for accurate object tracking in the video surveillance system. Traditional background generation techniques cause some problems when there have been non-moving objects for a long time. To overcome this problem, we propose a new background generation method using mean-shift and Fast Marching Method (FMM) that fuse pixel information along temporal and spatial dimensions. The mode of pixel value density along time axis is estimated by mean-shift algorithm and spatial information is evaluated by FMM, and then they are fused together to generate desirable background for non- moving objects for a certain period. Experimental results show that our proposed method is more efficient than the traditional method.