Traffic Sign Detection Based On Visual Attention Model

Traffic sign detection is a very important part of intelligent transportation and auxiliary driving system. It can help the driver enhance the safety by developing an algorithm of fast and accuracy detection and recognition of traffic signs. Also it can provide key technology for Intelligent Vehicle of enabling safe driving and autonomous navigation. So the task has gained great attention since recent years.Traffic signs are usually designed to specific color and shape so as to be separated mostly from natural and artificial background. Therefore the major method of traffic sign detection is combined with color segmentation and shape detection. The key of color based method is to choose an appropriate color space and then segment the image by means of threshold or color clustering. The shape base method is to determine whether the area is traffic sign through the geometry information. The mainstream methods are corner detection, edge detection, morphological filtering, template matching, and so on. The methods based on color have high requirement on the value of threshold interval.And it will make great difference on results because of color fading or illumination changing. The bottleneck of shape based methods is the geometric deformation due to the changing of observation angle.Aiming at these problems, we put forward some improvements. The main job of this paper is as follows:1、Based on the saliency tool model, We add the color-shape saliency map. Then it generates the obvious map with multiple features and finds the area that we are interest in.2、We use the improved color space and enhance the adaptability of threshold segmentation. It can weaken the influence of light changes and improve the segmentation accuracy at the same time.3、As to shape detection, we use the modified Shape Context and Chamfer Match methods. It can resolve the problem of scale of template and geometric deformation of different angles effectively.