License plate image data set generation method and system, server and medium

Abstract

The invention discloses a license plate image data set generation method and system, a server and a medium. The method comprises the following steps: generating a first synthetic license plate image and license plate text information according to a user demand and a system standard of an actual license plate; based on a real license plate image in the marked real vehicle image, migrating the style of the first synthetic license plate image by using any style migration based on a neural network to form a second synthetic license plate image; embedding the second synthetic license plate image into the real vehicle image according to the second synthetic license plate image and license plate rectangular frame coordinate information in the marked real vehicle image to form a first synthetic image and license plate position information; and finally, performing color conversion on the second synthesized license plate image in the first synthesized image to form a second synthesized image. Therefore, rich data of a real style can be flexibly and quickly provided for license plate detection and recognition, and the time cost and the labor cost required for obtaining a real license plate training image are greatly saved.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a method and system for generating a license plate image data set, a server and a medium. Background technique [0002] With the continuous application of license plate recognition technology in security monitoring, smart parking and other fields, an excellent license plate detection and recognition algorithm has become very important. The current mainstream license plate detection and recognition algorithm is developed based on deep learning, and the quality of the algorithm directly depends on the richness and diversity of the data set. [0003] The traditional method of establishing a data set relies on manually collecting vehicle images with license plates and labeling the license plate position information and text information in the images. The manual method is time-consuming and labor-intensive, and the collection process is likely to cause data imbalances in so...

AI Technical Summary

Problems solved by technology

The manual method is time-consuming and labor-intensive, and the collection process is likely to cause data imbalances in some areas where there are too many license plate data, and extreme weather such as rainy, snowy and cloudy days with less license plate data

Detrimental effect on the training of the license plate recognition model

[0004] On the one hand, the existing license plate data synthesis methods can only synthesize license plates that do not have a realistic style or the authenticity of the synthesized license plate is not strong

On the other hand, existing data enhancement methods in other fields are not universal and are not applicable to license plate data synthesis

Images