Limited scene-free license plate detection and classification method based on point guide positioning

Abstract

The invention belongs to the license plate detection and classification technology, and relates to a license plate detection and classification method based on point guiding positioning and without a limited scene, a license plate positioning method using point guiding is used, spatial information is explicitly utilized, the problem of insufficient utilization of the spatial information caused by using a full connection layer is avoided, the license plate spatial translation modeling capability is enhanced, and the license plate detection and classification efficiency is improved. Meanwhile, the positioning stability is improved through regression of multiple points; after the accurate position of the license plate is positioned, the features of the corresponding area are used to classify the license plate, and a classification regression task is separated into two channels, so that the contradiction of classification regression on feature translation is relieved, and license plate detection and classification in an unconstrained scene can be performed; the method can also be used for various inclined target detection tasks such as scene text detection and face detection, the detection precision reaches 98.5%, and the detection and classification effects are greatly improved.

Description

technical field [0001] The invention belongs to the license plate detection and classification technology, and relates to an unrestricted scene license plate detection and classification method based on point-guided positioning. Background technique [0002] With the rapid economic development, urban traffic congestion and other problems are becoming more and more serious, which puts forward higher requirements for the construction of intelligent transportation systems. As a key information processing technology in intelligent transportation, license plate detection plays a very important role in urban vehicle supervision. However, in practical applications, the license plate is easily deformed by the camera angle, so the license plate detection system still has the problem of low efficiency and accuracy. [0003] In recent years, with the advent of the era of big data and the improvement of computer computing power, deep learning has made major breakthroughs in the directio...

AI Technical Summary

Problems solved by technology

[0002] With the rapid development of the economy, urban traffic congestion and other problems are becoming more and more serious, which puts forward higher requirements for the construction of intelligent transportation systems. As a key information processing technology in intelligent transportation, license plate detection plays a very important role in the supervision of urban vehicles. , however, in practical applications, the license plate is easily deformed due to the influence of the camera angle, so the license plate detection system still has the problem of low efficiency and accuracy

[0003] In recent years, with the advent of the era of big data and the improvement of computer computing power, deep learning has made a major breakthrough in the direction of license plate recognition. The introduction of deep learning algorithms such as Faster R-CNN and YOLO has made the positioning and recognition of license plates new. Development, the general target detection can use the horizontal positioning frame to locate the rough position of the target very well, but the license plate is prone to distortion due to the angle of view of the camera. Using the horizontal frame to locate the license plate often includes the surrounding background information, which affects the classification of the license plate and recognition task

[0004] The license plate detection task requires the detector to estimate the accurate position of the license plate, not just provide a rough position of the license plate. In order to better describe the position of the license plate, the existing methods usually locate the compact bounding box of the license plate, that is, use the orientation box or positioning The four corners of the license plate. However, due to some serious defects, the existing methods cannot accurately locate the deformed license plate. First, the existing methods do not make full use of the spatial information. The commonly used regression-based method uses a fully connected layer The offset vector between the prediction and the anchor box regresses the bounding box of the deformed license plate. On the convolutional features of the model, each feature point responds to the adjacent area on the image. When using the fully connected layer to generate the vector, the feature The pixel-to-pixel mapping between the map and the image will be lost, so the lack of spatial information hinders the ability of the model to locate the target; secondly, the existing methods have the problem of contradictory checking and classification tasks, in order to achieve accurate bounding box positioning, The network should be able to respond to the subtle translation of the license plate to construct a translational equivariant feature; in terms of license plate classification, the feature map should maintain translation invariance with the target in space, and the previous method combines the two tasks in the same path for learning , this contradiction between target recognition and target positioning also brings difficulties to network optimization

[0005] It can be seen that for the license plate detection task, the existing methods still have the problems of insufficient utilization of spatial information and contradictions in detection and classification, which eventually lead to low detection accuracy of license plate. Therefore, more effective methods are urgently needed to enhance the utilization of spatial information and ease the detection and classification. the contradiction between

Images