Abstract: Incorporating region and feature information，this paper proposes a new algorithm for tracking vehicles under conditions of occlusion and splitting.Firstly the associations between motion regions are established by using the prediction information.Then making use of the result of region association，the method estimate whether the occlusion and splitting occur or not.The problem of occlusion is solved by tracking based on feature points，which are initialized according to region information and clustered by searching the matched points within a neighborhood of predicted position.The problem of blob splitting is handled by region matching after regional fusion in terms of some proposed rules.Experimental results show that the proposed method is characterized by good robustness and real-timeness.