Research On Traffic Sign Recognition Method In Complex Environment

Traffic sign recognition system,as an important part of advanced auxiliary driving system,not only ensures the safety of vehicles,but also improves the efficiency of vehicles.Traffic sign recognition system is mainly aimed at collecting traffic sign images under good environmental conditions.However,the driving environment of vehicles is more complex.Due to the dithering of vehicle camera,the occlusion of non-target objects and the interference of transmission media,the motion blurring,incompleteness and noise of the collected images are caused,which leads to traffic signs.The recognition accuracy is greatly reduced.In order to solve this problem,this paper studies traffic sign recognition algorithm in complex environment.The main research contents are as follows:(1)The algorithm model of SIOSVM is proposed and implemented.Combining the existing support vector machine model and online learning theory,aiming at the shortcomings of online support vector machine model,such as instability at the initial stage,learning efficiency decreasing with time and classification time increasing with time,an online recognition model of SIOSVM is proposed to classify and recognize the feature vectors of traffic signs after feature extraction.(2)A new improved convolution neural network model(improved CNN)is proposed and implemented.The improved CNN model is composed of CNN feature extraction model and SIOSVM classification and recognition model.Firstly,CNN feature extraction model is built through experimental analysis.Then,the extracted features are taken as samples and input into SIOSVM classification and recognition model.Finally,the improved CNN model is obtained through model training.(3)Verify the performance of the improved CNN network model.Six improved network models,namely CNN,OCNN,CNN,Le Net-5,Alex Net and HOG+SVM,were compared and tested.The performance of the improved CNN model was validated by using traffic sign images in complex environments as test samples.The experimental results show that the traffic sign recognition system based on the improved CNN network model has strong on-line adaptive ability;on the premise of guaranteeing the response speed,the model has better generalization ability than other classical models,and has higher recognition accuracy for traffic sign images in complex environments.