Vehicle Logo Recognition Method In Intelligent Transportation System

Vehicle logo recognition provides technical support for intelligent transportation systems,and plays an important role in efficiently and accurately identifying vehicles in intelligent transportation business applications.Vehicle logo images in actual production environments have influencing factors such as uneven illumination,tilted vehicle logos,stains and complex backgrounds.Vehicle logo recognition has always been a great challenge.The traditional vehicle logo recognition algorithm is mainly based on manual design features.The disadvantages are: the algorithm has limited anti-interference ability,low robustness,and cannot accurately and effectively locate and classify vehicle logos.Compared with manual design of features,neural network-based vehicle identification algorithms have great advantages in solving vehicle logo positioning and vehicle logo recognition problems in complex environments.Relying on a large amount of training data to obtain features is a necessary condition for neural network learning to ensure training,Verification,and detection of the accuracy of learning.This article conducts indepth research on the above issues.The main work is as follows:First of all,this article obtains 10321 original images of the bayonet monitoring system in the real environment from the actual work.The images have the characteristics of different lighting environment conditions,different camera resolutions,and the capture range covering single lane or multiple lanes,etc.,and effective vehicle logos are made in advance.Annotate,form training data,and use data enhancement methods to effectively amplify data samples to ensure the authenticity and adequacy of the algorithm data.Secondly,by analyzing the proportion of the car logo in the overall image and the characteristics of the car logo length and width ratio,using the SSD algorithm,setting up multi-scale feature maps and convolutional networks,effective model method training and calculation of effective target detection anchor points,Set 4-6 target recognition default boxes,pre-position the car logo,and accurately locate it through the regression algorithm.The YOLOV4 algorithm is used for target detection,and the backbone feature extraction network uses CSPDarknet53 and Mish activation function to enhance the feature network part using SPP and PANet to achieve stacking operation and feature fusion,which greatly improves the accuracy of target detection.Finally,through non-maximum suppression and evaluation algorithms,regression calculations are performed on the vehicle logo classification to ensure the accuracy and recall rate of the test results.For the unsuccessful positioning of the car logo image,the second intensive training is carried out to improve the generalization ability of the neural network to recognize the car logo.Under the condition of meeting real-time requirements,the extracted vehicle logo features are made more compact and robust.The average recognition accuracy of SSD detection vehicle logo reaches 92.8%,and the YOLOV4 vehicle logo algorithm vehicle logo recognition accuracy rate is 98.42%,in terms of accuracy.Significant improvement,greatly improving the practicability of the algorithm,providing favorable support for intelligent traffic vehicle logo recognition,and having broad application prospects.