Recognition of Pedestrian Traffic Light at Crosswalk for a Mobile Robot Using Deep Learning

Abstract

This paper describes recognition of pedestrian traffic light at crosswalk for a mobile robot using deep learning. In order for a robot to cross a crosswalk, the robot needs to recognize the color of the traffic light like a human. A recognition of traffic light by camera images based on manually designed image features is possible. However, it requires significant amount of labor to adjust parameters under changing lighting condition. Therefore, in this paper we tried to recognize traffic lights using deep learning. The proposed method consists of two processes: a detection of traffic light and a classification of a traffic light color using deep learning. These processes can be processed in an allowable time by a small computer mountable on a mobile robot. Through a series of experiments, the proposed method successfully recognizes traffic signal in real environments.