Hole-based traffic sign detection method for traffic signs with red rim

A traffic sign detection method is proposed in this paper based on detecting the inner nonred region of prohibitory and danger signs. It consists of five steps: red pixel extraction, red hole extraction, hole region filtering, semicircle combination, and final decision. In the step of red pixel extraction, a new red bitmap extraction method considering relative color of neighboring pixels is proposed to improve the accuracy of the extracted red bitmap. A series of weak classifiers are cascaded in the step of hole region filtering to quickly filter out the false alarms. A support vector machine is adopted in the final decision step to further reduce the false alarm ratio. Experimental results indicate that the proposed method is robust to many kinds of adverse situations including bad lighting condition, small rotation, out-of-plane rotation, similar background color, multiple signs clustered, and partial occlusion. Experiments on the GTSDB traffic sign dataset show that the proposed method achieves the recall of 99% and 97% for prohibitory and danger signs, respectively, while keeps the precision above 99%. In addition, the mean processing time of the proposed method is about 100 ms for each 1366 x 768 image from GTSDB on a Core I3 CPU, which confirms its suitability for real-time applications such as driver assistance systems.