Detection and recognition on traffic sign in complex scene

Abstract

Road traffic sign detection and recognition problem has always been a hot and difficult point in the field of intelligent driving. Complex road conditions always take challenges to the detection and identification of traffic signs. This paper attempts to use machine learning and deep learning algorithms to detect and identify traffic signs. The study is realized in two basic parts. The first part is detection on traffic sign. A color probability model based on YCbCr space and an improved color enhancement method based on MSER are proposed to detect Candidate Areas (CA). By using these two methods, most negative samples in CA can be reduced. Then, in the Region of Interest (ROI) feature extraction stage, PHOG features with strong spatial information expression ability are combined with SVM to classify and process. The second part is recognition on traffic sign. An improved lenet-5 model is used to promote speed and accuracy of the recognition. It has been proven that the proposed model has the characteristics of shorter classification time, higher classification accuracy and better generalization ability by comparing the other algorithm with ours used in GTSDB data set.