The Algorithm And System Application Of License Plate Recognition Based On Video Sequenc

License plate is an important feature of vehicle.License plate recognition is an important means of traffic surveillance and control.Although there are many algorithms for license plate video recognition,the accuracy and real time of license plate recognition are not satisfied,such as the license plate recognition scene in the toll lane,when the actual application scene,such as light interference,vehicle movement,shooting angle and small target.To this end,the paper,based on"deep learning based toll lane vehicle feature recognition system and Application Research",has carried out the research of license plate video recognition algorithm,and still has important theoretical significance and practical application value.This paper is based on the domestic and foreign status of license plate recognition,based on machine learning and deep learning mechanism,and is guided by the practical application of toll stations.A series of research work has been carried out with two algorithms of machine learning and deep learning of license plate recognition.In the study of machine learning recognition algorithm,the paper analyzes the principle of other algorithm structure,analyzes the problem of accuracy in the process of vehicle license plate location,license plate character segmentation and license plate character recognition,and gives the overall design scheme.In view of the recognition rate and applicability of the license plate recognition,the license plate area correction module is added.The paper also gives a solution to the shortcomings of machine learning license plate recognition,and gives the corresponding solutions,specifically:the problem of car license plate area detection is missing,the paper gives the multiple characteristics of the vehicle license plate to avoid the occurrence of leak detection;the location of the vehicle license plate is inclined and incompleted,therefore,the vehicle license plate area correction module is added to machine learning license plate recognition,and a detailed vehicle license area correction algorithm is given;vertical projection method can not divide the non connected domain characters.A self-adaptive vertical projection method is given in this paper,which can divide Chinese characters well.In the research of the algorithm for deep learning recognition,a data augmentation method is proposed for the dataset preparation module,which can automatically generate the augmented data calibration information.At the same time,the dataset is solved,and the data calibration is solved.For the license plate recognition module,the target detection algorithm of the RCNN series is analyzed;For the recognition of license plate characters of small targets,the optimization algorithm of deep network model SSD is proposed.The paper also gives some solutions to the problems of the deep learning of the license plate recognition,which is the problem that the data collection is often insufficient and the distribution of data is not balanced.A method of augmenting the data is proposed in this paper.The method can generate the calibration information of the corresponding supplemental data at the same time.In order to detect the small target,the paper improves the network structure of the SSD algorithm,so that it can detect and identify the small target license plate character,and its detection precision is 97.21%.The video recognition and detection of the actual scene of two algorithms:the deep learning license plate recognition algorithm is superior to the machine learning license plate recognition algorithm in terms of detection speed and recognition rate in addition to the need for GPU accelerated attachment conditions.The main innovation points and characteristics of this paper are:? In order to solve the problem that license plate recognition rate is not high in machine learning,the vehicle license plate area correction and vehicle license plate border elimination algorithm are given,thus the accuracy rate of license plate recognition is improved.? In order to solve the problem of inadequacy of dataset in deep learning and uneven distribution of data categories,a method of data augmentation is proposed to solve the problem of insufficient image data sets and additional image calibration.? Aiming at the problem of detecting small target difficultly by deep learning SSD algorithm,the SSD algorithm is improved,and the problem of small target detection and recognition is solved.