Zebra-crossing detection based on geometric feature and vertical vanishing point

Abstract

Zebra-crossing region detection from a zebra-crossing image is a challenging activity to support visually impaired navigating safely in outdoor environments. This paper proposes a zebra-crossing detection framework based on unique geometrical features of zebra-crossing. The unique geometrical feature is zebra-crossing stripe's edges are arranged in sorted order. Initially, pre-process the zebra-crossing images by applying the directional Gabor filter to eliminate the noisy and shadow's effect and improve the resolution of the zebra-crossing horizontal edges. Then Sobel edge detector is used to detect the horizontal edges. Potential horizontal edges are extracted by eliminating small and non-candidate edges. After that, the edge linking process is utilized to extract the longest horizontal edges. Extracted zebra-crossing edges are justified by arranging the horizontal edge segments in increasing parallel order. Finally, these parallel horizontal edges are used to measure the vanishing point to verify zebra-crossing edges from other analogous visual pattern and confirm the detection of zebra-crossing candidate region. Various zebra-crossing images are utilized to evaluate the proposed framework and presented outcomes demonstrate the adequacy.