ZebraRecognizer: Efficient and Precise Localization of Pedestrian Crossings

Autonomous mobility is a challenge for visually impaired people. In the last years, a number of solutions have been proposed in the scientific literature to support visually impaired people during road crossing. In our previous work we presented ZebraLocalizer a mobile application that detects zebra crossings and guides the user to safely cross. In this paper we present the ZebraRecognizer algorithm that improves the detection module of our solution and that applies innovative solutions in the field of zebra crossings recognition. The major contribution resides in the fact that ZebraRecognizer rectifies the ground plane hence removing the projection distortion of the extracted features. This leads to two major advantages: first, it possible to compute the relative distance between the user and the zebra crossing in meters. Second, the grouping and validation criteria specifically designed for the rectified line segments are much more effective, hence improving the accuracy of the recognition. An additional contribution consists in a significantly improved computation time. Indeed, ZebraRecognizer is 3 time faster than our previous solution, thanks to the adoption of a personalized version of the EDLines algorithm to detect the line segments.