This paper describes an accurate human silhouette extraction method as applied to video sequences. In computer vision applications that use a static camera, the background subtraction method is one of the most effective ways of extracting human silhouettes. However it is prone to errors so performance of silhouette-based gait and gesture recognition often decreases significantly. In this paper we propose two-step segmentation method: trimap estimation and fine segmentation using a graph cut. We first estimated foreground, background and unknown regions with an acceptable level of confidence. Then, the energy function was identified by focussing on the unknown region, and it was minimized via the graph cut method to achieve optimal segmentation. The proposed algorithm was evaluated with respect to ground truth data and it was shown to produce high quality human silhouettes.