Self-Similarity based Zebra-crossing Detection for Intelligent Vehicle

A zebra-crossing detection method for intelligent vehicle is proposed in this paper. The method is performed on a bird-eye view image called inverse perspective mapping image. The complete method includes two phrases. First, a morphological filer followed by horizontal projection is applied to fast extract candidate zebra-crossing regions, where the size and structure information of zebra-crossing are well utilized. Second, a recognition method which based on self-similarity is presented to identify the candidate regions. Given a seed region of a zebra-crossing, the recognition method can infer overall zebra-crossing region by matching and growing. Experiments on a great number of real images which consist of several challenge scenes demonstrate the effectiveness and efficiency of the proposed approach.