Fast Vision-Based Pedestrian Traffic Light Detection

Abstract

Detection of pedestrian traffic light is very important for the visually impaired. However, fast but accurate vision-based detection is not an easy task due to the complexity of background and illumination. In this paper, a fast vision-based detection system is designed. In the designed system, the background filter is applied to identify the candidate regions of pedestrian traffic lights. And the cascade classifier obtained by the Adaboost algorithm based on the multi-layer features is used to detect the pedestrian traffic lights. Testing results verifies the effectiveness of the designed system.