A Multi-layer Scene Model for Video Surveillance Applications

Foreground detection is the most important preprocess for video surveillance applications. However, classifying pixels of video frames into only background and foreground seems insufficient in real situations. In this study, we model the monitoring scene by using a multi-layer framework. The proposed scene model classifies pixels layer by layer into four different states, comprising background, moving foreground, static foreground and shadow. Different scenarios such as shadow elimination, abandoned object detection and object tracking were tested with the proposed scene model. The experimental results demonstrate it is quantified for real video surveillance applications.