Real Time Foreground-Background Segmentation Using Two-Layer Codebook Model

Abstract

Real-time segmentation of moving regions in image sequences taken from a static camera is a basic task typically used in surveillance system. This paper presents an improved codebook modeling strategy in real-time application avoiding high computational complexity and memory burden. The basic codebook algorithm is improved by constructing another layer to distinguish the type of the codeword and matching type status. Proposed algorithm can model periodic-like motion, hostile environment or illumination changes and stationary or abandoned/removed objects. Experimental results show that the proposed algorithm is effective, quick for motion detection, and can meet the demands of real-time applications on embedded hardware.