A license plate recognition method based on convolution neural network

Abstract

The invention discloses a license plate recognition method based on a convolution neural network. Firstly, the license plate image is preprocessed to construct the license plate training sample set; then the Faster R-CNN network model consisting of CNN convolution network layer, RPN candidate region extraction layer, ROI pooling layer and discriminant layer is constructed; continuous training through multitasking loss The model generates a Faster R-CNN network model with high accuracy. Finally, the trained Faster R-CNN network model is used to identify the license plate, obtain the location information and category information of the license plate character frame, and then use the prior knowledge to determine the position of the character, sorting and integrating the recognized license plate characters, and outputting the license plate recognition result. The method of the invention utilizes the convolution neural network to complete the license plate recognition, can complete the license plate recognition more quickly and accurately, and avoids the error accumulation problem caused by too many steps in the traditional algorithm.

Description

technical field [0001] The invention relates to the technical fields of image processing and computer vision recognition, in particular to a license plate recognition method based on a convolutional neural network. Background technique [0002] With the increasing number of urban vehicles, the pressure of traffic management is also increasing year by year, vehicle detection and recognition technology is still a hot spot that needs continuous innovation. For license plate detection and recognition, accurate recognition can be achieved in specific occasions. For example, in the gradually mature unmanned parking lot, the license plate recognition of the entering and exiting vehicles is carried out by means of image processing technology, and timing and self-service charging are performed. In the real environment of road traffic, the technical application of license plate detection and recognition is of great significance, such as road toll collection, traffic flow control, vehi...

AI Technical Summary

Problems solved by technology

The use of traditional algorithms has certain defects. Improper setting of the threshold value in the preprocessing of license plate characters may lead to low recognition accuracy; for modern license plates with various color categories, there are corresponding problems when using traditional image methods to distinguish license plate colors. Classification errors; license plate rotation adjustment, license plate character segmentation, and license plate character recognition also have errors. It is difficult to achieve a technical solution with high versatility and high accuracy when all errors are accumulated

Images