Semi-supervised Vehicle Recognition: An Approximate Region Constrained Approach

Semi-supervised learning attracts much concern because it can improve classification performance by using unlabeled examples. A novel semi-supervised classification algorithm SsL-ARC is proposed for real-time vehicle recognition. It makes use of the prior information of object vehicle moving trajectory as constraints to bootstrap the classifier in each iteration. Approximate region interval of trajectory are defined as constraints. Experiments on real world traffic surveillance videos are performed and the results verify that the proposed algorithm has the comparable performance to the state-of-the-art algorithms.