Robust foreground detection and shadow removing using local intensity ratio
Jinhai Xiang, Honghong Liao, Ning Wang, Heng Fan, Weiping Sun, Shengsheng Yu · 2013
Segmenting foreground objects from video sequences in real time is a fundamental step for video surveillance. In this paper, a local intensity ratio (LIR) model is proposed, which is robust to illumination change. And the distribution of the LIR is also discussed. Normalized local intensity ratio instead of pixel intensity is used in Gaussian Mixture Model (GMM) to segment the foreground without shadow. Afterwards, we use the ratio between foreground and image size, and contour orientation to fill foreground with holes. Experimental results show that the proposed method can get foreground objects in real-time, and eliminate more shadow compared to existing shadow detection methods.