A license plate recognition method based on deep learning

Abstract

The invention discloses a license plate recognition method based on deep learning, which comprises the following steps: step 1, training a deep convolutional neural network model M for license plate character detection; constructing a license plate character label set B; step 2, inputting the positioned license plate image I to a license plate character detection network M, and outputting a candidate license plate character set H; step 3, reordering the candidate license plate character set H obtained in the step 2 according to the abscissa hi.x of the upper left corner of the smallest circumscribed rectangle of the candidate license plate character from small to large to obtain a set C; step 4, further executing screening operation on the set C obtained in the step 3; step 6, sequentiallytraversing the set E obtained in the step 5; step 7, returning that license plate recognition result L obtained in the step 6. The method has the beneficial effects of effectively inhibiting the influence of license plate character adhesion, fracture, deformation, license plate dirt, license plate inclination, residual shadow on the license plate and the like.

Description

technical field [0001] The invention relates to the technical field of intelligent transportation, in particular to a deep learning-based license plate recognition method. Background technique [0002] In the past two decades, license plate recognition technology has greatly improved in recognition accuracy and algorithm efficiency. With the continuous advancement of technologies related to intelligent transportation systems, automatic license plate image recognition is considered to be a solved problem with mature solutions. Traffic flow analysis, vehicle speed measurement, and vehicle violation detection are representatives of many applications based on license plate recognition technology. However, in reality, there are many standard specifications of the license plate, the font and color of the license plate are obviously different, and the length of the characters of the license plate is different. In China alone, there are more than ten types of license plates, inclu...

AI Technical Summary

Problems solved by technology

[0009] To sum up, the current license plate recognition methods have the following deficiencies: 1) There are various license plate formats, and it is difficult for license plate recognition to take into account all types of license plates; 2) There are errors in methods based on segmentation (including traditional methods and methods based on deep learning) Cumulative problem, that is, the segmentation error generated in the segmentation stage will accumulate to the subsequent license plate character recognition stage, reducing the overall recognition rate; 3) Although the end-to-end license plate recognition network is designed to be able to handle multiple formats with variable character lengths License plates, but still can only deal with license plates with similar standards. The network suitable for single-line license plates cannot achieve ideal results on double-line license plates, and this type of network needs to be based on accurate license plate positioning, and the conditions are harsh; 4) Can be processed at the same time The license plate recognition methods of various license plate standards have high requirements on the clarity of license plate imaging, and it is difficult to successfully process low-resolution images; 5) Environmental factors (such as: light, noise, etc.) have a certain impact on the recognition of license plate images

Images