Technologies Of Real-time Traffic Sign Segmentation And Recognition

As economy is developing faster and faster nowadays, cars have been widely spread into millions of ordinary families. However, when cars give people convenience, problems of traffic congestion, traffic accident, road safeties and transport efficiency are becoming more and more serious, which bring great inconvenience to people's lives and works. In this case, researches of Intelligent Transportation System (ITS) are in progress and get more and more attention. Automatic Traffic Sign Recognition (TSR), as an important subtask of ITS, has been of great interest at home and abroad.As a necessary part of unmanned vehicles'vision system, real-time traffic sign detection and recognition has also become a hot research. Due to high requirements of real-time performance, the effect is not ideal in real environment and it remains to be a challenge.In the paper, we propose an effective method for traffic sign segmentation based on color distance in order to reduce the computational complexity during color space conversion and satisfy the real-time system. Through the selection of appropriate distance threshold on the basis of large number of scene image samples, it quickly obtains the binary image of the region of interest. Meanwhile, pyramid decomposition is used through the original image, so the segmentation can be done on the small image while recognition on the original image.In recognition stage, an adaptive threshold method is applied to get the binary blob image. In view of traffic sign's inherent geometric property, we extract ring projection feature and multi-scale global feature based on the binary image. At last, multi-template matching with feedback mechanism is used to classify the traffic signs, which improves efficiency and accuracy.The experimental result and test in scene show that our method can work well and achieve the traffic sign segmentation and recognition with sufficiently high processing speed and satisfactory accuracy. It can provide navigation information for the unmanned vehicles.