Research On Vehicle License Plate Detection And Recognition Technology Based On Neural Network

In recent years,with the popularization of the concept of intelligent transportation,license plate detection and recognition applications have spread across all aspects of our lives,including parking lot charging systems in communities,violation monitoring systems at intersections,and mobile handheld police systems for traffic police.With the advent of deep learning,the recognition accuracy of license plate detection and recognition algorithms based on deep learning has been further improved.At the same time,complex and changeable application scenarios put forward higher requirements for the accuracy of the algorithm.How to make the system work stably in a more complex environment has become the focus of research in recent years.This article first investigates related work at home and abroad,from the detection and recognition of license plates based on traditional image processing methods to the detection and recognition of license plates based on neural networks.After investigation,it is found that in a complex light environment,especially in an environment with insufficient light,the existing license plate detection algorithm has a low accuracy rate.To solve this problem,the existing work mainly uses image enhancement to solve the problem,first restore the picture to the state under the normal light environment,and then perform the detection and recognition processing.Although this improves the detection accuracy,it imposes an additional burden on hardware resources.Aiming at the above problems,this paper studies an adaptive dark light license plate detection algorithm based on deep learning.That is to say,automatic switching of the partial weights of the detection network for different light scenes,so that the network can maintain a high detection accuracy rate in normal light and dark light scenes.At the same time,some improvements have been made to the recognition network to make it easier to port to hardware.Then,considering the need for computing power of neural networks and the tendency of the license plate detection and recognition system to become terminal,this paper studies a configurable neural network acceleration hardware architecture based on the above algorithm.The design and implementation of modules such as convolution,pooling,bias,and network data processing have been completed.And by configuring two different parallel modes,the efficient use of the multiplier is realized.On this basis,this paper implements and experimentally analyzes the proposed algorithm and hardware.Through experiments,it is found that the adaptive dark-light license plate detection algorithm proposed in this paper is 6.2% higher than the original model in the dark-light image data set.At the same time,the hardware design has an average data error of 4.3% with the algorithm under INT8 accuracy.And in the case of a working frequency of 100 MHz,a frame rate of 22 frames per second is reached.Finally,this article summarizes all the work,analyzes the deficiencies of algorithm design and hardware design,and looks forward to subsequent improvements.