License plate recognition method in unconstrained scene

Abstract

The invention provides a license plate recognition method in an unconstrained scene. The method comprises the following steps: S1, carrying out deblurring processing on an input license plate image by using a deblurring network designed in multiple stages to obtain a clear image; s2, inputting the obtained clear image into a vehicle detection network, and positioning a vehicle bounding box; step S3, zooming the vehicle bounding box; s4, inputting the vehicle bounding box image into a license plate inclination detection network, positioning a license plate and performing inclination correction; s5, the corrected license plate image is sent to an OCR network, character segmentation and character recognition are carried out, and a license plate number is obtained; according to the method, more complex fuzzy and inclined license plate pictures can be processed, and the recognition accuracy in a complex scene is effectively improved.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a license plate recognition method in an unconstrained scene. Background technique [0002] With the rapid development of the global economy and the continuous deepening of the urbanization process, the number of motor vehicles has continued to explode. No matter in developed countries or in developing countries, the road traffic construction of the governments of various countries will never keep up with the growth rate of motor vehicles. Therefore, traffic problems also follow. One of the inevitable ways to solve the existing traffic problems is to build an efficient license plate recognition system. For any car, the license plate number is the unique identification of the vehicle's identity. Therefore, the license plate recognition system is widely used in intelligent parking lot charging, searching for vehicles, managing traffic roads, etc. [0003] License plat...

AI Technical Summary

Problems solved by technology

[0006] Disadvantage 1 of the existing technology: the license plate recognition method generally consists of three steps: license plate detection, character segmentation, and character recognition

Directly detect the license plate, the accuracy will be affected

[0007] Disadvantage 2 of the existing technology: In image deblurring, most multi-stage design methods use encoder and decoder structures, which can extract rich context information, but do not retain enough spatial texture details

[0008] Disadvantage 3 of the existing technology: In image deblurring, most multi-stage design methods directly transfer the output of the previous stage to the input of the next stage, so the obtained effect is not very good

[0009] Disadvantage 4 of the existing technology: In the problem of license plate image tilt, the algorithm network layer based on deep convolutional neural network is too deep, resulting in excessive network load and affecting the overall license plate recognition speed

Images