Moving shadow detection with support vector domain description in the color ratios space

Moving shadow detection is a fundamental step in video-surveillance applications since it is generally confused with foreground. We propose a novel statistical non-parametric method to detect moving shadow in a road traffic image sequences. We consider a diagonal model to describe the shadow distortion in the RGB color space. A support vector domain description (SVDD) algorithm is applied in the color ratios space in order to discriminate shaded pixels from foreground.