Scene segmentation based on seeded region growing for foreground detection

This paper proposes a scene segmentation method for foreground detection in outdoor visual surveillance. An outdoor scene is divided into three parts: road, sky, and other region by using seeded region growing (SRG) algorithm in which road region gets initial seeds by using the motion information of vehicles and sky region gets initial seeds by using a probabilistic classifier. As a result, instead of detecting foreground objects in a conventional pixel by pixel manner, detection can be performed selectively on the pixels of partial regions in terms of different surveillant requirement so as to facilitate and accelerate foreground detection. Experimental results show the contribution of our proposed method.