A geometrical feature point-based approach for tracking a certain region of the extended target is presented. In this approach, the geometrical feature point is used as the template for tracking, which is similar to template matching. Combining the multi-scale Harris detector with Scale Invariant Feature Transform (SIFT) descriptor, these feature pionts can be detected and described. A new weighted similarity measure and updating strategy is adapted in order to improve the accuracy of point matching. Test results show that the proposed algorithm can track the extended target accurately and robustly even though the pose of the target changes significantly between frames.