Abstract
This paper proposes a reliable vision method to detect traffic signals and to recognize their status. The goals of the method are to recognize the traffic signals robustly at any hour of the day or night and to achieve a detection range of more than 100 m. The proposed method detects illuminating signal lights primarily by using color information and circluar shape. Various cues, such as shape-based signal detection and knowledge of the illumination order of traffic lights and the 3D position of the traffic signals, are integrated to remove false positives from the detected candidates. A quantitative evaluation using image sequences captured on urban roads by an on-board color camera proved the effectiveness of the proposed system.
Keywords: Traffic signal recognition, color image, on-board camera