Research On Vehicle Logo Recognition Method Based On Deep Convolutional Neural Networ

Intelligent traffic,an indispensable part of the wisdom of the city,is closely related to people’s lives.In view of the fact that the road traffic system can’t identify the vehicle identity through the license plate,the researchers put forward the new vehicle identification method and technology based on logo.However,the actual environment and the structure of the vehicle,the traditional vehicle logo recognition method can’t be more efficient to solve the problem.In this paper,the structure of the vehicle combine with the depth of convolution neural network superior learning and expression,a new method of vehicle marking recognition based on deep convolution neural network is proposed.The main contents of this thesis include the following four aspects:1.Establishment of common vehicle standard library as experimental data set.Experimental data set include Volkswagen,Toyota,Honda,Audi car four kinds of car image,in different weather,different lighting,different camera angle,different colors,different proportions.90%of each of the vehicle picture is for the training set,training network model,debugging network model parameters.The remaining 10%is a test set to verify the accuracy of the network model for vehicle identification.2.Research on the location of the car face and feature extraction method.Achieving the positioning of the car face by the method of Adaboost algorithm.Using SIFT and SURF invariant features to analyze and compare the face of the car,putting forward to the improvement method of the location of the car.Effectively improve the car face feature points in the face image information in the proportion,so as to improve theoretically the positioning accuracy of the car face.3.Research on method of vehicle logo location based on heat bar.After analysis of the texture characteristics of the heat sink.Using the horizontal gradient projection and vertical gradient projection to segment the heat sink to locate the vehicle logo location.Compared with the direct method of vehicle logo location face,improve accuracy.4.Study on vehicle logo recognition method based on convolutional neural network.Lenet,Alexnet,Googlenet three kinds of depth convolution neural network model is used to identify the four kinds of common logo in the actual road,and the accuracy rate of up to 95%.Compared with the traditional SIFT and SURF vehicle logo recognition method based on the invariant feature and BP neural network method,the accuracy and efficiency improved significantly.