Design Of License Plate Recognition System Based On Image Processing And Deep Learning

Since the 21 st century,the number of vehicles in China has exploded,and now China has become the country with the highest number and growth of vehicles.The following traffic management problems are also greatly increased,and the license plate number,as the car’s identity card,contains all the information of the license plate.Therefore,license plate recognition has become the top priority of daily traffic management.In this paper,for the license plate recognition system,digital image processing and deep learning are used to carry out the license plate detection and recognition experiments respectively,and finally select the appropriate method to complete the design of the license plate recognition system.In the license plate detection based on image processing,a small hole filling algorithm is proposed to eliminate the holes in the license plate area caused by morphological operations.At the same time,an interactive ratio detection algorithm based on color mask is proposed to improve the detection effect.Aiming at the problems in license plate character segmentation,a license plate region optimization algorithm is proposed to eliminate some image noises and extract effective regions.Aiming at the low precision of character segmentation,a character filtering algorithm is proposed to remove the non character images.The license plate detection experiment is carried out using four algorithms: license plate detection based on image processing and image color mask interaction ratio algorithm,fast rcnn,ssd,yolov4 tiny.The experimental results show that the license plate detection algorithm based on yolov4 tiny has the best effect.The license plate recognition experiment is carried out by using four algorithms based on improved character segmentation and template matching,improved character segmentation and SVM,improved character segmentation and CNN,LPRNet.The experimental results show that the LPRNet based license plate recognition algorithm achieves the best results.Based on the above experimental results,the license plate detection algorithm based on yolov4 tiny and the license plate recognition algorithm based on LPRNet are selected to form the license plate recognition system.Based on the license plate detection algorithm selected above based on yolov4-tiny,combined with the difficulties of license plate detection in the current night environment,a data enhancement algorithm based on the retinex theory is designed.yolov4-tiny network is selected in the experiment,and the algorithm is used as the data enhancement strategy for training.The final results show that the algorithm has achieved the expected effect of data enhancement,expanded the data set and improved the detection accuracy.