Robust Background Maintenance for Dynamic Scenes with Global Intensity Level Changes

In background subtraction, the maintenance of significant backgrounds is very critical under various scene changes. In this paper, we propose a background maintenance method for dynamic scenes including global intensity level changes caused by the changes of illumination conditions and camera settings. If global intensity level changes abruptly, the conventional background model cannot discriminate true foreground pixels from the background. The proposed method immediately modifies the background model based on the estimation of level changes by mean-shift. Saturated pixels are handled by an additional scheme. In real-world experiments for dynamic scenes, our proposed method outperforms previous methods by adaptive background maintenance and handling of saturated pixels.