Research On Key Technology Of Character Recognition For Vehicle Detection

As an important part to ensure driving safety,vehicle inspection is increasingly heavy because of the increasing number of the motor vehicles.In addition to the quality and safety detection,the vehicle identification information such as license plate number and frame number manually identified and recorded by the people,which always play an important role in tracing the vehicle detection records in the future.With the development of Deep Learning technology in the past decade,a breakthrough has been made in the target detection,optical character recognition and other fields.The network model based on Deep Learning has achieved or even exceeded the accuracy of human in many data sets.This shows that Deep Learning technology has the ability to solve practical problems,so it is the general trend and the focus of our attention to use new technology in practical projects such as text recognition in vehicle detection,and to replace manual operation within the scope of our ability,so as to reduce the workload of relevant staff.The research work of this paper mainly includes the following five aspects:1.In this paper,we use the prior knowledge that the length of the target text is fixed and the space between characters is equal,and design a leak module to improve the recognition accuracy of the model for long text.2.Inspired by the related research of "deceptive Deep Learning",this paper uses the method of reducing or adding noise to process the images to change the original numerical distribution and improve the accuracy of character classification.3.During the completion of this project,our laboratory collected and marked several kinds of images including license plate and frame number from the vehicle management office,and made relevant data sets to verify the feasibility of this method.4.A vehicle inspection text recognition system has been designed and implemented,which is written in Python.And a simple interface is developed by pyqt5 for achievement display.At the same time,we also used Django framework to provide web services,and support to call the system functions in LAN.