Advanced crosswalk detection for the Bionic Eyeglass

One of the functions of the Bionic Eyeglass, a mobile navigation and orientation device, is to inform blind and visually impaired individuals whether a pedestrian crosswalk is located in front of them. In this paper, we present an improved algorithm to detect and recognize pedestrian crosswalks by using enhanced color preprocessing through mean shift segmentation. Detection of crosswalks is carried out by a partially adaptive CNN algorithm that also considers the perspective distortion of the objects to be detected. Using the new segmentation method we get much better estimations for the road surface area, which results in better overall recognition rates.