Traffic Sign Segmentation and Recognition in Scene Images

Traffic Signs provide drivers with very valuable information about the road, in order to make driving safer and easier. They are designed to be easily recognized by human drivers mainly because their color and shape are very different from natural environments. Automatic traffic sign detection and recognition is important in the development of unmanned vehicles, and is expected to provide information on road signs and guide vehicles during driving. This paper deals with traffic sign detection and recognition from image sequences. In order to reduce the computational complexity in the scene image processing, an effective method for traffic sign segmentation based on color distance is proposed in this work. The scene image is mapped to a matrix by computing the color distance. Through the selection of appropriate distance threshold on the basis of large number of scene image samples, it quickly obtains the binary image. To obtain better classification performance, we use linear support vector machine with the Distance to Border features of the segmented blobs to get the shape information, and then realize the rough classification based on Color-Geometric Model. Traffic sign classification is implemented using RBF kernel based support vector machine with edge related pixels of interest as the feature. The experimental result shows that our method can work well and achieve the traffic sign segmentation and recognition with sufficiently high processing speed and satisfactory accuracy.