Research On License Plate Location And Recognition Based On YOLO Deep Learning Algorithm

Due to the defects of traditional license plate recognition algorithms,it is difficult to accurately locate and recognize the license plate under the circumstances of poor image quality and poor shooting Angle,especially in the parking lot and toll booth.With the increase of private cars,traffic jams,parking difficulties,environmental pollution,road safety and other problems have seriously affected people’s daily life.The society urgently needs a more efficient intelligent traffic management system,in which the management of license plate information is the key to determining the success of the traffic management system.Based on this,this paper proposes efficient recognition and accurate positioning of license plates based on YOLO deep learning algorithm.The research contents are as follows:(1)License plate localization.The paper analyzes the traditional algorithm based on edge feature in the license plate location,license plate location based on color feature and learning algorithm of license plate localization based on morphological characteristics,and depth of license plate location algorithm based on convolution neural network positioning principle and existing defects,choose the depth study YOLOv3.1 model of the improved algorithm for experimental research.Through the research,it is found that:when locating the same license plate at the same time,the response speed is several times higher than the traditional algorithm.After 60,000 iterations of training,the accuracy can be maintained at about 99.2%,which is higher than most deep learning algorithms that have been applied in life.(2)For license plate character recognition,this paper analyzes the operation process and deficiencies of license plate character recognition algorithm based on template matching,license plate character recognition algorithm based on feature statistics,and AlexNet character recognition algorithm based on convolutional neural network.Chose200000 sample data sets,with the method of control variables comparing experiment,by comparing the findings: YOLOv3 algorithm character recognition performance good AlexNet convolutional neural network model of the characters of high accuracy of1.131%,numbers and letters of recognition accuracy was 1.649% higher,faster response time in this experiment to select sample size,less than 1 millisecond.(3)In the experiment of YOLOv3 localization and recognition,it was found that the improved YOLOv3.1 presented better performance.After 60,000 iterations,the accuracy of license plate localization could reach 99.2%.The best data in the experiment reached99.98%,but it was not very stable.It can be predicted that the YOLO model of deep learning has great potential in the field of license plate positioning and recognition,and further research is needed in order to achieve ultra-high precision positioning and rapid and efficient recognition,which plays an important role in the construction of intelligent traffic.