License plate correction method and device and video collection device

Abstract

The invention discloses a license plate correction method and device and a video collection device. The license plate correction method comprises the following steps that: obtaining an image to be detected and a target detection and identification model based on a convolutional neural network; through the target detection and identification model based on the convolutional neural network, detecting four pieces of angular point information of a target license plate in the image to be detected, wherein the angular point information comprises a square area, an angular point category and the confidence coefficient of the angular point; and according to four pieces of angular point information of the target license plate, correcting the target license plate in the image to be detected. When the technical scheme of the invention is adopted, various deformations which are performed on a plane figure since the license plate rotates in a three-dimensional space can be accurately descried, the deformed license plate in the three-dimensional space can be corrected through the four pieces of angular point information, and therefore, the method is slightly influenced by the change of a license plate number, and has the advantages of being high in detection rate, low in false detection rate, high in angular point category identification rate and accurate in positioning.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a license plate correction method and device and a video acquisition device. Background technique [0002] With the continuous improvement of economic level and people's living standards, vehicles in large, medium and small cities are constantly increasing. At the same time, the ensuing traffic problems have attracted more and more attention. In the process of identifying traffic violations, the automatic and accurate identification of illegal vehicle license plates is particularly important. However, in the license plate recognition process of illegal vehicles, it is first necessary to correct the tilt direction of the license plate in the image to be detected. At present, the more commonly used method for analyzing the tilt direction is to analyze the tilt direction of the characters through principal component analysis (PCA) and other methods, and determine the tilt...

AI Technical Summary

Problems solved by technology

[0003] However, in the actual process of determining the inclination direction of the license plate, it will be interfered by some inclined characters in the license plate, thereby affecting the judgment of the inclination direction of the license plate. For example, if there are multiple "7"s in the license plate, it is easy to cause misjudgment of the inclination direction of the license plate; Moreover, determining the tilt direction of the license plate in the prior art is usually limited to dealing with the rotation of the license plate in the camera image plane, and is invalid for any rotation of the license plate in three-dimensional space.

For example, the perspective deformation of the license plate in the three-dimensional space will cause the characters to produce a visual effect of large and small characters. The method of determining the tilt direction of the license plate in the prior art cannot realize the correction of this type of license plate.

In addition, the recognition algorithm of license plate tilt angle in the prior art does not have scale invariance. For example, there are requirements on the size of characters. If the size of characters changes greatly, the accuracy of the algorithm will be greatly affected.

Images