License Plate Recognition Terminal System In Resource-Constrained Environment

The deep learning based license plate recognition system has become a research topic for many scholars.The deep learning based license plate recognition method has high accuracy and robustness,but is completely dependent on the amount of data learned and the performance of the execution device.In today’s Io T era,the concept of cloud-side end has been proposed and has received wide attention and application.By deploying license plate recognition algorithms on terminal devices,faster and more accurate license plate recognition can be achieved,while also reducing the computational pressure on the cloud and network bandwidth usage,and its portability makes the devices more widely used in a wide range of application scenarios.Based on this,this thesis makes a detailed research and analysis on the above related technologies,and designs and implements a license plate recognition system with a handheld terminal as a carrier,integrating all the functions of computing,storage and network communication on the terminal device,which is mainly developed by combining manual feature methods with deep learning.After experiments,it is proved that this method can perform satisfactorily on a low-performance terminal device without a data set.The license plate recognition system designed in this thesis implements the following functions on the handheld terminal: license plate recognition,network communication,human-computer interaction and intelligent display.The main work of this thesis is as follows:1.A license plate detection method based on the combination of color features and texture features of license plates is proposed.The method can locate two two-line electric bicycle number plates and two one-line motor vehicle number plates,a total of four types of number plates without any difference.For the license plate color features,the HSV color threshold range is redetermined,and the license plate is initially located according to the character color of the plate and the cross area of the plate base color.Then some a priori knowledge of the license plate is used to further screen out the candidate regions and obtain the license plate location information.Due to the variety of interference factors in the natural environment,contour detection and contrast adjustment methods are used as supplementary means in the image pre-processing stage.Finally,according to the different cases of skew or distortion produced by different categories of vehicle license plates,Hough(Hough)transform and perspective transform are used to make targeted corrections,so as to obtain license plates in a better state.2.The adaptive threshold de-bordering algorithm based on the projection method is proposed.Firstly,the image is binarized according to an improved binarization method to obtain a license plate image with more complete features.Then the license plate image is projected,its pixel distribution position is analyzed,and the segmentation threshold is calculated adaptively by the created mathematical model to remove the license plate border.The method is applicable to different kinds of vehicle license plates and can cope with the presence of complex borders,and has excellent segmentation performance especially for electric bicycle license plates with complex interference around the plate.After obtaining a clean character region,the characters are extracted by projection method again.3.Character recognition model based on convolutional neural network.In order to adapt to the low performance handheld devices,this thesis proposes a character recognition network with only eight layers.The extracted character images are fed into the network for recognition.4.Multi-task parallel processing.Due to the limited performance of the handheld device,parallel programming is used in the development in order to be able to improve the system processing efficiency.Through the process pool,a total of four processes are opened to process the tasks simultaneously.At the same time,a logical system for inter-process communication was designed so that information could be passed between processes in a smooth and orderly manner.Through experiments,the method improves the system processing efficiency and fully mobilizes the hardware resources of the device.5.Network communication.In this thesis,the information interaction between the handheld device on the external network and the server on the internal network is realized through the 4G network module and TCP/IP protocol.The handheld device sends the license plate number to the local server,and the server transmits back the corresponding vehicle and owner’s details.