Real-time Background Modeling/Subtraction using Two-Layer Codebook Model

Abstract

Abstract—In this paper, a real-time method for foreground-background segmentation is proposed. This method extracts structure of background and models it in layered codebook. Layered codebook is a simple data structure containing two codebooks that is defined per pixel. The first layer is main codebook, other is cache codebook, and both contain some codewords relative to a pixel. Main codebook models the current background images and cache codebook is used to model new background images during input sequence. Initially main and cache codebooks are empty. Main codebook is constructed during train sequence. During input sequence, foreground-background is segmented and two-layered codebook is updated. Proposed algorithm can model moving backgrounds, multi backgrounds and illumination changes and this is efficient in both memory and computational complexity.