Cutting-free end-to-end license plate recognition method

Abstract

The invention provides a cutting-free end-to-end license plate recognition method. The method comprises the following steps: S1, collecting a license plate recognition data set, and constructing a training set and a test set for training a recognition network; S2, designing a feature extraction module for extracting license plate character features; and extracting features of the license plate characters and distribution features of the characters in license plates; S3, designing a deconvolution module for recovering blurred license plates and further optimizing expressions of license plate character features; S4, designing an output module of a recognition network by using the feature extraction network and a deconvolution module; S5, training the recognition network by using training setlabels; when the recognition network outputs the corresponding probability vector containing the character probability, obtaining final license plate numbers by using a greedy algorithm; according tothe invention, the recognition process of the license plate numbers can be completed only by using the convolutional neural network, and the invention has the features of no character cutting, end-to-end recognition, no license plate correction, rapidness and light weight.

Description

technical field [0001] The invention relates to the technical field of license plate recognition in intelligent traffic control, in particular to a cutting-free end-to-end license plate recognition method. Background technique [0002] License plate number recognition technology (license plate recognition) is an important part of the intelligent traffic management system. By identifying the license plate number of the vehicle, the vehicle information can be uniquely determined. License plate number recognition technology has a wide range of application backgrounds, such as parking lot management system, community vehicle access management, campus vehicle access management, etc. Further, with the gradual use of security monitoring systems, the demand for license plate number recognition in any scene will inevitably be further improved. [0003] Although the license plate recognition technology has been applied in real life, the existing license plate recognition technology i...

AI Technical Summary

Problems solved by technology

This new solution certainly solves the problem of traditional character cutting, but the cyclic neural network usually uses the context information of the text to improve the recognition rate when recognizing, but the characters of the license plate number are randomly generated and have no contextual connection, resulting in Recurrent Neural Networks May Actually Learn Wrong Text Information

At the same time, the training and calculation of the cyclic neural network are more complicated than the convolutional neural network.

In addition, these methods usually have a license plate correction module in the process of realizing the license plate recognition, which makes it easier to recognize the license plate number on the originally rotated and twisted license plate, but the cost of correcting the license plate is usually relatively large, and the recognition rate will be slow. dramatically drop

Images