An efficient shadow removal method using HSV color space for video surveillance

An approach to detect and remove cast shadows of moving objects was proposed in this paper. Gaussian mixture model with only one learning rate a was used for background subtraction and modeling. Initial classification of foreground pixels into object pixels and shadow pixels were performed using saturation property of HSV color space. In the hue difference or brightness ratio based shadow detection step, mixture of two Gaussian density functions were used to model the density function computed on the values of hue difference or brightness ratio. Expectation Maximization (EM) algorithm was used to estimate the Gaussian parameters. Threshold calculations were based on estimated parameters used to obtain set of shadow pixels. Local region property based shadow detection step uses local brightness ratio property to obtain the set of shadow pixels. Results of experiments performed on different scenarios shows that the proposed approach is robust and accurate.