Key Technologies Of License Plate Recognition Based On Deep Learning

The license plate recognition technology refers to a technique that can acquire the vehicle video on the road,and automatically extract the license plate image from the vehicle license and then perform image recognition.The key of license plate recognition system is license plate recognition algorithm.The traditional license plate recognition algorithm divides the entire license plate recognition process into four modules: license plate location,license plate extraction,character segmentation,and character recognition.The advantage of this algorithm design is that the algorithm can be designed separately for each module,thereby reducing the overall algorithm design complexity.However,the disadvantage is that the accumulation of errors between modules can easily lead to a reduction in the accuracy of the final license plate recognition.Based on this,this paper uses the deep learning technology to improve the traditional license plate recognition algorithm.The number of modules has been decreased from four modules to two parts which including license plate detection and license plate recognition.In the license plate detection part,based on the investigation of the local and foreign license plate location algorithm,this thesis analyzes the advantages and disadvantages of several common license plate location algorithms and a license plate positioning method based on color information and edge information is designed.Combined with the two-class model SVM(Support Vector Machine),the identified license plates are screened to improve the accuracy of license plate positioning.After experimental tests,the license plate location algorithm designed in this thesis has a higher positioning accuracy.In order to better compare the traditional license plate recognition algorithm and the undivided license plate recognition algorithm designed in this thesis,this article specifically implements the character segmentation link of the traditional license plate recognition algorithm.Firstly,the license plate is corrected by Radon transform-based tilt correction algorithm.Then use vertical projection analysis method to character segment license plate.Finally,using the improved template matching method to identify the license plate characters.Experimental results show that the recognition accuracy of the traditional license plate recognition algorithm depends heavily on the accuracy oflicense plate character segmentation,which greatly reduces the robustness of the algorithm.Based on the research of deep learning technology,this thesis applies the convolutional neural network in the license plate recognition algorithm to realize the undivided recognition of license plate characters.Based on the study of various classical convolutional neural networks,the classical convolutional neural network model AlexNet was selected and modified for the specific technical requirements of license plate recognition.Train the model and apply it to the character recognition module in the license plate recognition algorithm.After a lot of test data verification,the license plate recognition algorithm proposed in this paper has a higher recognition accuracy in a complex environment and has certain practical significance.