Efficient recognition of speed limit signs

Abstract

An automatic traffic sign detection system would be important in a driver assistance system. In this paper, an approach for detecting Norwegian speed limit signs is pro- posed. It consists of three major steps: Color-based filtering, locating sign(s) in an image and detection of numbers on the sign. About 91% correct recognition is achieved for a selection of 198 images. I. INTRODUCTION Being able to recognize traffic signs would probably be important for vehicle safety systems in the future. Such systems could assist drivers on signs they did not notice before passing them. Specifically speed limit sign recog- nition - studied in this paper, could inform drivers about the present speed limit as well as giving an alert if a car is driven faster than the speed limit. In the future, autonomous vehicles would probably have to be controlled by automatic road sign recognition. Recognizing road images have been studied for a long time. The first known attempt for making a real-time system was by Akatsuka and Imai (1). Many techniques have been proposed since then. The standard technique for detecting and recognizing road signs consists of three steps (2). First, color segmenta- tion or color thresholding is applied to emphasize possible signs in an image. Thus, restricting the search area in the image. Second, template matching is applied for shape detection. Third, specific signs are detected using template matching or neural networks. A color image is often transformed from the RGB (Red, Green, Blue) color space into the HSV (Hue, Saturation, Value) color space. Color segmentation then becomes easier (by applying it on the hue value only rather than the three RGB values). The hue value is invariant for the illumination as well. However, the hue is not suited for grey-level segmentation since it has a constant value along the "grey- level" axis. The value will be unstable near this axis too. Small perturbations in the RGB signals may cause large variations in the hue. A traffic sign detection system consisting of four stages has been proposed by Wei et al (3). First, a color-based filtering is applied to filter out image regions that have color characteristics similar to the color found in one of the signs known to the system. That is, the color input image is converted into a binary image using the filter. Second, the boundaries of the regions are smoothed by applying close