An Efficient Traffic Signs Recognition Method for Autonomous Vehicle

An efficient traffic signs recognition (TSR) method is presented to solve the problems such as the poor real-time performance and low accuracy of existing methods in the intelligent transportation system (ITS). Firstly, some image areas are selected according to experiments, which are preprocessed to adapt to different environments, and are split into four channel images, i.e. red, blue, yellow and black. Then, the qualified contours are selected from the outer contours of each channel image, and the convex hull processing for those contours is conducted for the second selection. Next, the circle and square contours are selected according to their characteristics such as areas, perimeters and Hu invariant moments, and their internal images are obtained as regions of interest (ROIs) from the original high resolution image. Finally, each ROI image is matched with templates through histogram scaling and translation matching (HSTM algorithm) by using horizontal and vertical histogram characteristics, and the optimal matching result is regarded as the final recognition result. In Chinese Intelligent Vehicle Challenge, the autonomous vehicle equipped with the proposed TSR system has recognized all the specified signs, whose recognition rate is up to 95% and recognition speed is up to 8Hz～10Hz. The proposed method proves its advantages in real-time performance and in accuracy compared with other existed methods.