A Zebra-Crossing Detection Algorithm for Intelligent Vehicles

This paper describes a robust algorithm for detecting zebra crossings to provide information for intelligent vehicles’ autonomous navigation in urban environment. First, transform images framed by the camera to the top view by an Inverse Perspective Mapping (IPM); extract the region of interest (ROI) from IPM image by local threshold segmentation. Then, extract datum band from RIO by analysis of every region’s length, direction, as well as mutual relations between the bands; finally, extract all bands belonging to the zebra crossing. The zebra crossing’s distance and direction are easily calculated in the IPM image. Experiments from 6000 street scenes with and without crossings demonstrate the robustness of the proposed algorithm.