Ellipsoidal Method for Shadow Detection Based on Normalized RGB Values

This paper presents a new approach of foreground and shadow segmentation in visual surveillance environment. Coordinate systems are firstly defined based on normalized RGB values, and then an ellipsoid is constructed according to scattering feature of shadow pixels. Consequently, moving shadow detection is treated as a clustering problem in above coordinate systems, and all the pixels inside the ellipsoid are extracted as potential shadow pixels. Finally, post-processing is further performed by using the geometrical properties of shadow to verify and confirm pixels classified as shadows in previous step. Experimental results validate the algorithm on a suite of indoor and outdoor video sequences. The performance of the method is markedly higher than that of the two well-known moving shadow detection methods, especially under changing illumination conditions.