Research And Application Of Vehicle Recognition And License Plate Recognition Based On Deep Learning

With the increasing of vehicles in China,traffic congestion and traffic safety problems are becoming more and more serious,so the demand for intelligent management of traffic and parking lots is also growing.In this intelligent management system,the vehicle model and license plate number are similar to the "ID card" of the vehicle,so it is of great practical significance to study how to accurately identify the vehicle model and license plate.Therefore,this paper studies the accurate vehicle recognition and license plate recognition,and develops the human-computer interaction interface between them,so as to facilitate the promotion.Firstly,this paper introduces the theories related to deep learning.Then,four convolutional neural network models which is Alexnet,VGG-16,Resnet-18 and Inception v3 respectively is constructed in MATLAB and the structure and parameters of the four models is analyzed one by one.For vehicle recognition,in order to solve the problem that coarse-grained vehicle recognition can not meet the demand of intelligent management system nowadays,this paper uses the vehicle type database of Stanford University for experiments,so as to realize the identification of vehicle brand,model and year.In order to avoid the phenomenon where the model is difficult to converge due to the large number of parameters in the training process,four models is trained through transfer learning for feature extraction training in this paper.In addition,all four models use the Softmax classifier for their classifiers.Finally,the trained model is tested,and the results show that Resnet-18 has the best test performance with the testing accuracy of 96.12%.In license plate recognition,for the problem that camera,environment and other external factors may affect the brightness and shade of the license plate image,morphologicalprocessing methods with different parameters for the brighter and darker images is used in this paper.Then,the Radon algorithm is used to correct the possible tilt of the license plate image in this paper.Then the character segmentation is carried out according to the projection law of the binary license plate image.Finally,support vector machine is used for character recognition to obtain the license plate number of the testing license plate image.In order to facilitate the using of intelligent traffic and parking management system managers,the human-computer interaction interface for vehicle recognition and license plate recognition is designed and tested in this paper.The results show that the human-computer interaction interfaces have good human-computer interaction,where the Resnet-18 model after training is used for vehicle recognition.Finally,this paper summarizes the vehicle recognition,license plate recognition and human-computer interaction interface,and looks forward to model,database,the limitations of license plate recognition and the future research direction.