Research Of Vehicle Logo Detection Algorithm Based On Deep Learning

With the rapid development of vehicle manufacturing technology,transportation efficiency has been greatly improved.However,the popularization of vehicles has brought safety,regulation,and management problems while improving traffic efficiency.The construction of intelligent transportation system is an important measure to solve this series of transportation problems.Vehicle logo detection is an indispensable part of the intelligent transportation system.In recent years,many vehicle logo detection methods have been proposed,and the field of vehicle logo detection has developed rapidly.However,the existing vehicle logo detection methods still have the following three problems.Firstly,existing vehicle logo detection methods have low accuracy.Secondly,existing vehicle logo detection methods have not solved the impact of motion blur on vehicle logo detection task.Thirdly,existing vehicle logo detection methods do not solve the background interference problem in the vehicle logo detection task.Aiming at the problem one: the existing vehicle logo detection methods have low accuracy.This paper proposes a vehicle logo detection method based on deep learning.The method consists of two parts: EOC clustering algorithm and VL-YOLO detection algorithm.The EOC algorithm can effectively eliminate outliers in the dataset and perform unsupervised clustering.After EOC clusters the dataset,the priori boxes will be obtained,and the priori boxes will be provided to VL-YOLO.The VL-YOLO algorithm optimizes the network structure and combines deep and shallow feature maps of different sizes through feature fusion.On this basis,VL-YOLO built a multi-scale detection network structure to achieve accurate vehicle logo detection.We carried out a series of comparative experiments.Experimental results show that the vehicle logo detection method proposed in this paper achieves 98.1% detection accuracy.Aiming at the problem two: the existing methods have not solved the influence of motion blur on the vehicle logo detection task.This paper proposes a vehicle logo image deblurring method named Filter-DeblurGAN.The Filter-DeblurGAN acts on the image deblurring stage of the vehicle logo detection task.This method has a good effect on improving the accuracy of vehicle logo detection under the condition of motion blur.First,Filter-DeblurGAN evaluates the quality of the image and judges whether the image has motion blur problems.Then,the method performs shunt processing on the images according to the identification result,and then repairs the selected images with motion blur problem.Finally,Filter-DeblurGAN outputs the repaired images.We conducted multiple experiments to demonstrate the effectiveness of Filter-DeblurGAN,and the results show that the method has good deblurring performance.In addition,Filter-DeblurGAN can improve the accuracy of vehicle logo detection under motion blur conditions.Aiming at the problem three: the existing vehicle logo detection methods have not solved the background interference in the vehicle logo detection task.This paper proposes a method of extracting the region of interest of the vehicle logo based on the relevance of the vehicle structure.Due to the complexity of the background,the problem of false detection is unavoidable.Therefore,the reliability of the vehicle logo detection results will also be affected.The method proposed in this paper uses the position correlation between the vehicle lights and the vehicle logo to extract the region of interest of the vehicle logo.Experiments show that this method can effectively reduce the impact of background interference on vehicle logo detection task.This paper demonstrates the effectiveness of the above methods through a large number of experiments.Experimental results show that the proposed methods can improve the detection accuracy of vehicle logo,effectively solve the influence of motion blur on vehicle logo detection task,and reduce the interference caused by complex background.Part of the results of this paper have been published in SCI journals.