Robust Background Subtraction Using Geodesic Active Contours in ICA Subspace for Video Surveillance Applications

Abstract

Current background subtraction methods require background modeling to handle dynamic backgrounds. The purpose of our study is to investigate a background template subtraction method to detect foreground objects in the presence of background variations. The method uses a single reference image but the change detection process allows change in the background including illumination changes and dynamic scenes. Using indoor and outdoor scenes, we compare our method to the best state-of-the art algorithms using both quantitative and qualitative evaluation. The results show that our method is in general more accurate and more effective.