Research And Implementation On Fast Traffic Sign Detection And Recognition

As the development of car industry, the research is aimed at increasing driving satety and automation of the vehicle navigation process. In order to free the people from driving,the researchers worldwide have studyed this problem for amlost 30 years. However,in the complicated natural scenes, the problems of different lighting condition, partial occlusion and the efficiency of the algorithm make the research of traffic sign detection and recognition for manless driving far from mature.This paper fisrt illustrates the research status of this problem, and analyses the advantages and disadvantages of the major methods.The research of this paper is aimed at improving the efficiency of traffic sign detection and recognition,and implements the real-time traffic sign recognition systerm.A fast traffic sign detection algorithm using the typical color information of traffic signs is proposed.The input images and videos are enhanced with color constancy algorithm to reduce the influence of illumination on the detection performace. Color and morphological feature is used to complete coarse detection, and the fine detection is completed through the SVM based on HOG feature.A traffic sign recognition method is proposed based on multi-feature fusion.The algorithm fist classifies traffic signs into several super classes, then gets the final classification result at the second stage.The paper respectively researches and compares the accuracy of traffic signs recognition based on HOG feature,LBP feature and Gabor feature. Experimental results show that the proposed algorithm outperforms above three methods and the recognition speed is fast.At last,the paper introduces the TI Davinci platform of hardware and software architecture.The code is ontimized by usins hardware resource.