Research On Traffic Sign Detection And Recognition Algorithm

With the rapid development of economy,the concept of intelligent transportation is becoming more and more popular, and intelligent car has gradually become a part of our daily life. Traffic signs detection and recognition system is the key point of Intelligent Transportation Systems (ITS) research. However, the intelligent vehicle traffic sign recognition system needs to recognize signs under any natural environment condition,and the result would be influenced by various external interference conditions.Nowadays,the traffic signs detection and recognition process mainly include the issues as following. First, in the process of color segmentation of traffic signs detection, the performances of segmentation related to the light condition, and the light intensity would affect the segmentation accuracy. Second, in the natural environment, the traffic signs may be occluded by tree branches or buildings in front of the signs, which may result to get incomplete contour of signs. Third, projection distortions would occur if the image plane is not parallel to the signs, such as circular signs are recognized as oval signs, rectangle signs are recognized as quadrilateral signs. Fourth, because the traffic signs detection and recognition system is applied to the driving vehicles,it is necessary to have good real-time performance for the algorithm.Aiming at the problems of traffic signs detection and recognition,we propose a traffic signs detection method by combine the color and shape information based on color segmentation and contour analysis. This method is divided into three phases: color segmentation, contour analysis, and real-time recognition. In the color segmentation phase,histogram equalization algorithm is used to preprocess the original image to enhance the contrast, and we use the frequency tuning algorithm to get the saliency map. Then we can obtain the traffic sign regions of interest by binary processing. In the contour analysis phase,we filter out the interference information according to features of contour length and aspect ratio. Then convex hull of candidate contours are extracted to calculate normalized Fourier Descriptors (FD). Finally,we get detection results by compared with standard data. In the real-time recognition phase, Support Vector Machine is used to classify the traffic signs based on Histograms of Oriented Gradients (HOG) features. We trained various classifiers according to the different categories of traffic signs, and this method will improve the recognition rate to 98.79%. In order to objectively evaluate the presented approach, a large number of traffic sign images under different environment were used as test images.Experimental results demonstrate that the proposed method is robust enough against occlusion, projection distortion, different weather and illumination conditions. The detection and recognition rate is higher than the other existed methods with better real-time performance.