Block-Based Hough Transform for Recognition of Zebra Crossing in Natural Scene Images

Abstract

In this paper, a block-based Hough transform is proposed to recognize the zebra crossing in natural scene images. Overlapping blocks are laid on the region of interest (ROI) in each image. For each patch in the block, two processes are performed successively. First, preprocessing is adopted for edge detection, whereas the adaptive thresholding is used to minimize the effect of various shadows. Second, parallel lines detection is adapted to recognize the zebra crossing, whereas the Hough transform is used for straight lines detection. When all the blocks are processed, the angles of parallel lines are averaged to provide the direction of the zebra crossing, and the accumulative scores are synthesized to provide the position of the zebra crossing. The performance of the proposed method is evaluated by testing results based on numerous images.