Research On Vehicle Detection And Vehicle Type Recognition Based On Deep Learning

With the improvement of people's living standard,the number of cars is increasing,which brings more traffic problems.Traffic problems can be effectively alleviated by vehicle detection and vehicle type recognition technology.Compared with traditional machine learning and image processing methods,deep learning has a very prominent advantage in the field of computer vision.Therefore,this paper adopts the method of deep learning to solve the problems of vehicle detection and vehicle type recognition in traffic,and provides technical support for intelligent transportation system to alleviate traffic congestion and other issues.Specific research contents are as follows:Firstly,the theoretical basis of convolutional neural network in deep learning is expounded,including convolutional neural network structure and network training method.among which,convolution layer,pooling layer,several common activation functions,fully connected layer,classifier and regularization be emphatically studied in network structure,the back propagation algorithm and the gradient descent method are mainly studied in the network training.Secondly,the traditional vehicle detection method cannot adapt to the changing traffic environment.This paper uses the convolutional neural network to adaptively extract vehicle characteristics.Researching and analysing the region-based convolutional neural network series algorithm and selecting the R-FCN framework with good performance as the vehicle detection network.In order to further improve the accuracy of vehicle detection,this paper improves the shared convolutional network in the R-FCN framework by increasing the number of shared convolutional layers and ultimately increasing the average accuracy of vehicle detection by more than 4%.Thirdly,the traditional vehicle identification is only a rough classification of vehicles.This paper will classify the models,including 256 models of 72 brands.Research and analysis of commonly used image classification networks,finally,DenseNet with excellent classification ability is selected as the vehicle classification network,and verify the effectiveness of the vehicle classification network in the CompCars and VehicleID.Finally,traffic images and videos in real scenes are used to verify the effectiveness of vehicle detection network and vehicle classification network.The results show that the vehicle detection network can achieve high detection accuracy in different environments.At the same time,the vehicle classification network can achieve good classification results for vehicle types in traffic video.