An Imagery based Automatic Traffic Sign Detection Method for Sri Lankan Road System.

Abstract

Traffic signs are very useful for drivers to drive more safely and easily in different driving conditions. Inaccurate tr!ffic sign detection and misclassification due to driver ignorance and bad driving conditions result adverse impacts in driving environments. In this paper, we propose an automatic traffic sign detection method to detect danger warning signs and speed limit signs in Sri Lanka. The effects of occlusion or partial occlusion of traffic signs, weatherworn deterioration of traffic signs, variation in illumination and variation in perspective are considered. The proposed method uses the background colours of red, yellow and white of Sri Lankan traffic signs to find the regions of interests. The incoming frames of a real time video captured by a video camera in a moving vehicle is first normalized to enhance the pixel values of red and yellow colour regions. Then, the candidate regions are detected using maximally stable extremal regions (MSERs) algorithm. Python based Open CV library is used for the implementation of the proposed detection method. We compare the detection performance of our method and popular Hue-Saturation­Value (HSV) colour transformation method. The results show that the proposed approach outperforms the HSV colour transformation method. The detection rate of the proposed method shows higher detection rate and it is insensitive to the variations of lighting conditions and weather conditions.