Red and Yellow Traffic Lights Detection Robust to Various Lighting Conditions

Abstract

A traffic light is one of the most important traffic signs for drivers to drive smoothly and safely. However, drivers may fail to pay attention to it by some chance. Automatic traffic light detection may be a great help to cover driver’s carelessness. This paper presents an automated traffic light detection method that works robustly under varying lighting conditions. We focus on the detection of red and yellow traffic lights because those signs are more crucial than green traffic lights for avoiding traffic accidents. The proposed method utilizes solely the color information in the CIELab color model, excluding the intensity information. The method is based on a simplified fast radial symmetry transform (FRST). The FRST is a fast implementation of the circular Hough transform. We also introduce a shape descriptor called solidity to reduce false traffic light detections. Simulation results show that the proposed method can significantly improve the precision and recall (sensitivity), that is, 79.19% and 87.5%, respectively, compared with conventional approaches.

Keywords: Traffic light detection; Fast Radial Symmetry; Computer Vision