Background initialization for the task of video-surveillance

Abstract

In this work, we propose a region-wise and batch processing approach for background initialization in video-surveillance based on a spatio-temporal analysis. First, the related work has been explored. Then, the efforts are focused on developing a new background initialization approach to outperform the literature performance. To this end, a temporal analysis and a spatial analysis are performed. In the first stage, we use a previous work techniques adding motion information to increase performance. In the second stage, a multipath iterative reconstruction scheme is performed to build the true background under the assumption of background smoothness, i.e. the empty scene is smoother than the scene with foreground regions. Finally, the results over challenging video-surveillance sequences show the quality of the proposed approach against related work.