Parametric tracking of human contour by exploiting intelligent edge

We adapt the idea of intelligent scissors for contour tracking in dynamic image sequence. Tracking contour of human can therefore be converted to tracking seed points in images by making use of the properties of the optimal path (intelligent edge). The main advantage of the approach is that it can handle correctly occlusions that occur frequently when human is moving. Non-uniform rational B-spline (NURBS) is used to represent parametrically the contour that we want to track. In order to track robustly the contour in images, similarity and compatibility measurements of the edge are computed as the weighting functions of optimal estimator. To reduce dramatically the computational load, an efficient method for extracting the region that we are interested in is proposed. Experiments show that our approach works robustly with sequences with frequent occlusions.