Research On Traffic Sign Recognition Method In Auxiliary Driving System

Intelligent transportation system has been emerged with the explosive growth of car in various countries.The driver assistant system and traffic sign recognition system,as a significant part of the intelligent transportation system,have incited ongoing and growing research interest.However,there are still many technical problems need to be solved for identifying the traffic signs accurately.In this paper,the detection and recognition of the traffic signs are analyzed as follow:This paper presents a traffic sign detecting method based on the search with large and small windows simultaneously.The two sliding windows are used to scan and extract the HOG features in the traffic scene image,and two SVM classifiers are used to judge whether the traffic signs is exists in the window.Finally,the result is obtained by processing the results of the two groups.The proposed method shows the ability to resist the false detection while improving the detection accuracy effectively,and to detect the traffic signs in most of the adverse conditions.This paper also presents a traffic sign recognition method based on stepwise refinement.The method exacts the HOG features according to the characteristics of the traffic signs,such as colors and shapes.The traffic signs are roughly classified into several categories by using the primary SVM classifier,and then refine each category step-by-step by the secondary SVM classifier until the final recognition result is obtained.For the two categories of prohibition and warning signs,the classification accuracy has improved significantly by extracting and refining the red outer edge of the region.Numerical simulation and experiment show the feasibility and the validity of the proposed method.In the last part,based on the generalization characteristic of deep learning and sensitive characteristic of target recognition,a deep network model is proposed by constructing and training the convolution neural network under Convolutional Architecture for Fast Feature Embedding(Caffe).The proposed model has ability to detect and recognize the traffic signs simultaneously.The experimental results show that this method can significantly reduce the missed and false detection of traffic signs.