Research, Based On The Saliency Map Of Traffic Signs Detection Method

Traffic sign recognition(TSR) system is part of driver assistance and intelligent transportation systems. TSR has a very important value and a wide range of market demand. Because of the complexity of the scene and the diversity of traffic signs, the traffic signs automatic detection and recognition level cannot meet practical application requirements. We generate color and edge character based on the detection of saliency map, then search the candidate traffic signs area by the classification of pixels. Finally we use local hough transform method to search the position of traffic signs area and remove the flase targets.Traffic sign is set to show, so its design and installation process has fully considered the visual characteristics. In this paper we analyze the saliency characteristics based on the visual attention mechanism which is remarkable from natural scene. And improve the calculation method of traffic sign saliency map due to the problem from classical Itti model.WTA mechanism is often used to detect visual attention focus which is a fixed radius circle, and the circle is larger than the actual target area. This paper designs a target-searching method based on clustering principle. First remove weak significant figure features area through a global threshold then classify the pixels in dynamic way according to their characteristic value and positions. Divide and combine these clusters according to pixels'coordinates variance in cluster and the distance between each other and each area will contain a single traffic sign at last. The experiment results show that the method can select candidate areas of traffic signs in relatively short time.The principle of saliency model makes it unable to search the accurate location of the target area, we use Hough transform method to detect the pure position of the traffic sign in local space and remove the flase target according to its shape. Elliptic detection is using their symmetry testing in different directions of the axis. Analyse the lines'positions, intersection angles and points to detect triangles and rectangulars. The experimental results show that the proposed method generates good results.