Vehicle Detection And Vehicle Type Classification Based On Deep Learning

Vehicle detection technology is an important part of the intelligent transportation system and widely used in various traffic scenarios.Whether it is assisted driving,traffic flow detection,or Electronic Toll Collection System,it plays a major role.Furthermore,Electronic Toll Collection System charges according to different vehicle models,this article studies the image-based vehicle detection and vehicle classification technology.Firstly,the vehicle detection method based on license plate location was studied,that position the vehicle area according to license plate location and vehicle symmetry.The contrast analysis is based on the license plate location methods such as color,character and edge.The experimental results show that the character-based license plate positioning method achieves the best positioning accuracy of 92.58%.After that,the vehicle front face area is located based on the license plate position and vehicle symmetry,and the vehicle is finally constructed.A vehicle face partial image set is built.Secondly,the classification method of machine learning vehicle models based on combined features is studied.The classification performances of HOG features,Gabor features,HOG+Gabor+PCA combined dimensionality reduction features,and HOG+Gabor+LDA combined dimension reduction features are compared and analyzed.The results show that the combine feature of real-time performance and accuracy has better performance than single features.Under comprehensive contrast,the support vector machine based on Gaussian kernel has the best classification performance when extracting the HOG+Gabor+LDA combined dimension reduction features.The method achieves the accuracy of 89.56%.Thirdly,four deep learning vehicle classification method,which are based on existing excellent network and transfer learning,are studied.The four models are VGG-16-based vehicle classification model(VGG16-VCM),InceptionV3-based vehicle classification model(InceptionV3-VCM),and Xception-based vehicle classification model(Xception-VCM)and Resnet50-based vehicle classification model(Resnet50-VCM).The experimental results show that the InceptionV3-VCM model has high precision,training accuracy is as high as 96.48%,verification accuracy is as high as 83.85%.Finally,a fused deep neural network model for vehicle classification(FDNN-VCM)is proposed.The fused model consists of an input layer,an intermediate layer,a fusion layer,and an output layer.The intermediate layer integrates three deep neural networks,InceptionV3,Xception,and Resnet50,and they are contactend in parallel.The confusion matrix,recall,precision,accuracy and boxplot of the three models were compared through experiments.The experimental results show that the vehicle classification method of the deep network fusion model(FDNN-VCM)performance is far better than the traditional model,the classification accuracy is as high as 95.43%.