Robust Detection of Abandoned Objects Based on Modified Dual Background Segmentation

Abstract

A robust real-time method for detection of abandoned objects in complex outdoor environments is presented. The system introduces a dual background segmentation method based on feedback modification. One background updates fast and the other updates slow. The slow background is then modified by the detection result to exclude the static foreground. They contribute to build the static foreground region masks which are used to update a counter. Static foreground regions are then extracted from the counter and further filtered by a light and shadow eliminator. The filtered regions are then classified into abandoned or removed objects by comparing the feature of its surroundings and the interior boundary vicinity of the fast background and slow background. The algorithm is simple and computationally less intensive. Experimental results on public video datasets and real image sequences acquired outdoor show its practicality and robustness to moving vegetation, shaking camera, illumination change and crowed scenes.