Crosswalks recognition through CNNs for the bionic camera: Manual vs. automatic design

Although accessible pedestrian signals are more and more frequent at crosswalks in busy intersections, visually impaired people will not be able to get about independently until the infrastructure reaches full coverage on the routes they are using. In this paper, we present algorithms to detect and recognize pedestrian crosswalks developed in the framework of the research to create a Bionic Eyeglass, a mobile navigation and orientation device for blind and visually impaired people. We compare the results of manually designed algorithms with ones automatically generated via Genetic Programming.