Traffic Sign Detection And Pattern Recognition Based On Branch Tree Support Vector Machines

Traffic signs recognition is a hotspot in today's field of computer science. With the rapid development of economic and society,the traffic problems have become increasingly prominent.That makes pepole search for a new resolution to traffic jam and safety.Intelligent transportation system synthesizes machine vision,graphic technology, the computer and communication technology that supplies a new method for traffic problems. It applies to the automatic driving system and offers real-time traffic information,so it gradually develops into a new science that have a wide perspective and practicability.And more and more scholars pay close attention to it.Traffic sign's automatic detection and recognition is part of ITS,which can realize real–time traffic information collection,meanwhile analyze and process the information to feed back to the driver to keep safe driving. There are two stages in the common methods of automatic traffic signs detection and recognition:the first detection stage is to locate the regions that contain traffic signs. The second stage is recognition of traffic signs. This stage feature extraction and classification is carried out.The traffic signs have some common features,such as fresh color and particular geometric shapes ,so the detection and recognition of traffic signs generally based on color and shape information.However,because of the sun,shading,complicated background the traffic signs become fuzziness and distortion.That affects the accuracy of traffic signs recognition.This paper presents an automatic road sign detection and recognition system based on binary tree SVM. Color based segmentation techniques are employed for traffic sign detection. The coordinates position of traffic sign in images used for shape classification are obtained by orthogonal projection. An algorithm based on Hough transform was proposed to achieve better shape classification performance.Recognition of traffic signs are implemented using branch tree multi- classifer SVM with geometry semantic feature as the feature vector.