Traffic Sign Recognition Research Based On Convolutional Neural Network

Traffic signs can be seen everywhere in modern society,it also plays an important role at the same time,as a kind of information carrier to riding worker,it pass various instructions.In the current rapid development of intelligent transportation,a growing number of cars equipped with intelligent recognition system on traffic signs,the intelligent recognition system can help the people of driving reduce cognitive effort to recognize traffic information,more important is to reduce hidden danger from human cognitive deviation,as a result,more safety and reliable traffic signs recognition system is the driver's demand,it is also a research focus of the auto makers.However the images taken by actual automobile which are running,hard to avoid appearing image distortion,fuzzy phenomenon,in addition to these,there exist interference of external uncontrollable factors,such as bad weather conditions and so on,which makes the study of traffic sign recognition system is facing many difficulties,and the actual applications are far from mature stage.This study through the extensive literature search,found in the literature review convolution neural network is applied in many aspects,such as image recognition,speech analysis has made significant achievements.Convolution carried a multi-layer neural network device,can effectively reduce the image distortion when recognize the image and the scaling factors of interference,but its deep structural takes too long when made in identifying images,have no way to achieve rapid response in the actual driving,not suitable for the application environment of the real-time demand.According to the reality,this study attempts to design a optimized convolutional neural network,mainly to solve on the basis of the guarantee accuracy increase its practical usefulness.The main work of this paper include the following aspects: in order to avoid long time-consuming problem in the application of traditional convolution neural network,an optimized convolution of neural network algorithm is proposed.On the issue of traffic signs detection,through combine the method of color and shape,first step is using the SVM color classifier converting the original image to gray image,using shape template matching to extract the regions of interest,interest areas be further refining subsequent,then input to the optimized convolutional neural network to detection,detection algorithm in Germany Traffic Signs Detection Benchmark to verify the feasibility,results show that the algorithm has high detection accuracy,and to light and shade,or rotating has strong robustness.According to traffic signs recognition problem,presents a hierarchical classification algorithm.Traffic signs can be divided into several categories: first,and then according to the characteristics of all kinds of logo targeted as the image preprocessing,after processing of image input optimized convolution neural network for fine classification that specific category,finally proposed recognition algorithm with traffic signs recognition based on Germany Traffic Signs Recognition Benchmak and compared with other algorithms.Results show that the proposed algorithm not only assure the high classification accuracy,but also greatly improves the operation speed,more suitable for real-time demand higher traffic sign recognition system.