Research And Application Of Convolutional Neural Network In Vehicle License Plate Recognition Of Multi-channel Network Video Stream

With the rapid development of society and the significant improvement of people’s living standards,the number of cars is increasing rapidly.In order to protect the safe travel of citizens and facilitate the supervision of road traffic,a powerful video surveillance system has been built in the market.However,there are relatively few researches on license plate recognition systems that can be directly applied to existing surveillance videos.Therefore,research on real-time recognition of license plates for video streams has certain development prospects.This paper applies a target recognition algorithm YOLO9000 based on the convolutional neural network to design a multi-channel network video stream license plate intelligent recognition system,which can be directly applied to existing urban surveillance video,for toll system,traffic monitoring,parking lot management.Such practical applications have significant practical value.Compared with the traditional license plate recognition algorithm,the study uses the deep learning method to reduce the preprocessing of input images,character feature extraction and other steps,and allows the sample data to have a certain degree of defects,achieving the detection of multi-scale images,small targets and Multi-target.The improved YOLO9000 network model can realize license plate detection and character detection at the same time.It has good fault tolerance,parallel processing and self-learning ability.It has advantages for the complex and variable license plate recognition problem in the application environment,and solves the problem of slow speed of traditional methods.After testing,the accuracy of license plate detection,character detection and character recognition reached 98.9%,97.16% and 97% respectively,which has a good recognition effect.The license plate recognition system designed in this paper firstly applies VLC to play the multi-channel network video stream distributed by the IP camera,performs real-time transmission and decoding,and extracts Keyframes containing the completed license plate information;Then,the multi-threaded concurrent processing method is used to call the intelligent plate recognition algorithm to process the image queue to complete the license plate detection,character detection and character recognition in sequence;Then,the multi-threaded concurrent processing method is used to call the intelligent recognition algorithm to process the image queue to complete the license plate detection,character detection and character recognition in sequence;Finally,the real-time display of the license plate recognition result and the video stream data is realized,At the same time,the statistical analysis of the license plate recognition result and the accompanying time information is completed,and the statistical result is saved as log information,realizing the real-time,efficient and stable license plate information extraction of the multi-channel network video stream.