License Plate Recognition Algorithm Based On Deep Learning In Complex Scenes

With the rapid development of society,intelligent transportation plays a more and more important role in real life.As an integral part of intelligent transportation system,license plate recognition technology is particularly important.The existing license plate recognition technology is more accurate for license plate recognition in ideal environments such as relatively fixed license plate position and good light conditions.However,there is still room for improvement in license plate recognition in complex scenes such as abnormal weather,complex background and license plate tilt.In order to improve the performance of license plate recognition in complex scenes,a license plate detection and recognition method based on deep learning is proposed and integrated into an intelligent license plate recognition system.The main contents of this dissertation are as follows:(1)An improved license plate detection algorithm based on YOLOv5 is proposed.In order to improve the accuracy and speed of license plate detection in complex scenes such as complex background and abnormal illumination,the improved algorithm based on YOLOv5 locates the license plate position,and introduces the attention mechanism into the original algorithm to strengthen the feature extraction ability and retain more valuable feature information.The improved YOLOv5 algorithm is used to detect license plates and corners,which lays the foundation for subsequent license plate correction and character recognition.The experimental results show that compared with classical target detection algorithms such as Faster R-CNN and SSD,the improved license plate detection method based on yolov5 has better performance for license plate detection in complex scenes.(2)An end-to-end license plate character recognition algorithm is proposed.This paper designs an LPCRNet algorithm for end-to-end recognition of license plate characters based on convolution recurrent neural network.The algorithm skips the process of character segmentation,takes the whole license plate image as the input,and directly recognizes the license plate characters in the image through the trained model.LPCRNet algorithm corrects the inclined license plate at the input,sends the corrected license plate image into the joint network of convolutional neural network and cyclic neural network to learn the sequence characteristics of the license plate and predict its label,and finally transcribes the predicted label with CTC method to obtain the recognition result of license plate characters.Experiments show that LPCRNet algorithm is better than other character recognition methods in license plate character recognition under the condition of tilt and uneven illumination.(3)An intelligent license plate recognition system based on deep learning algorithm is designed and implemented.The intelligent license plate recognition system integrates two modules:license plate detection and license plate character recognition.It loads the improved license plate detection algorithm based on YOLOv5 and the LPCRNet algorithm for end-to-end license plate character recognition to complete the license plate recognition task,and designs a visual page to display the results of license plate detection,license plate correction and character recognition in the recognition process.The test results show that the intelligent license plate recognition system has good robustness and real-time performance for license plate recognition in a variety of scenes.Figure 33 table 5 reference 75...