The Research Of Vehicle Logo Recognition Method Based On Gradient And Anti-Blur Features

Vehicle logo recognition(VLR)is an important part of the intelligent transportation system(ITS).In recent years,researches on the VLR have been increasing,but most methods have not fully and comprehensively analyzed the characteristics of vehicle logo images.On one hand,the descriptive ability of features extracted are limited and can not fully describe the characteristics of vehicle logo images;On the other hand,the extracted features are too complicated and high-dimensional,which increases the speed of VLR,and influences their application in real-time systems.The variety of vehicle logo shapes,the interference of ambient light,the low resolution of the vehicle logo images and the unstable image quality make the VLR very challenging.In order to improve the recognition rate of VLR,this paper proposes two VLR methods: VLR based on enhanced localization of edge gradient features and VLR based on anti-blur feature extraction.The main work of this paper is as follows:(1)By summarizing the characteristics of the vehicle logo images,this paper points out the shortcomings in the existing VLR methods,which help us to make further research plans for the VLR.(2)VLR based on the local quantization of the enhanced edge gradient feature is proposed: this method extracts the enhanced edge feature information analyzing the overall characteristics of the vehicle logo images.Then,through the local quantization process of the features,the number of features is reduced while learning the key features.A large number of experimental results and comparisons on the HFUT-VL1 and XMU vehicle logo datasets verify the effectiveness and robustness of the method.(3)VLR based on anti-blurred features extraction is proposed: this method is to solve the problem that the blurred vehicle logo images existing in the VLR.Firstly,the vehicle logo image pyramid is constructed,and then the anti-texture blurred feature and anti-edge blurred feature of the vehicle logo image are extracted.The experimental results on the blurred vehicle logo dataset show that compared with other methods,the VLR method based on anti-blur feature extraction can effectively identify the blurred vehicle logo image,and also achieves high recognition rate on the HFUT-VL1 dataset.