Fast background subtraction algorithm by statistical background updating

摘要

As the first step of many machine vision tasks,background subtraction is required to be simple and fast.In this paper,we propose a fast background subtraction algorithm,which aims to speed up conventional algorithm by calculating distributions of foreground both in the horizontal and vertical directions instead ofconventional pixel-wise methods while updating the background.The algorithm in this paper is mainly an improvement of the multi-frame average background updating algorithm,and we use the ordinary background subtraction to obtain foreground objects.For high-resolution videos,we can use background areaswhich are not updatedfor further acceleratinginpost-processing step of background subtraction.