Detection of pedestrian crossing from focus to spread

In order to single out pedestrian crossing from real-life scenarios, this paper does the priori and likelihood modeling in Bayesian framework based on the defined block-based Markov random field. Furthermore, using coarse to fine technique and covariance matrix descriptor, it is focused on the crossing's central position by maximizing a posterior probability. Finally, it spreads around this center and determines the crossing's scope by randomly generating some rectangles. Experimental results illustrate its effectiveness in real applications.