Research On Roadside Parking Management Technology Based On Deep Learning

In recent years,due to the growth of car ownership,the original parking lot has been unable to meet the growing demand for parking.The implementation of roadside parking can relieve the problem of parking space shortage.It can make full use of road resources and relieve the problem of parking space shortage at a lower costAt present,many roadside parking management systems need to carry out road reconstruction The introduction of hardware facilities such as geomagnetic detector for detection and identification will consume a lot of manpower and material resources,and may affect the normal driving of other vehicles.With the popularization of urban surveillance,it is easier and easier to obtain surveillance video,thus the video-based roadside parking management scheme has become a feasible schemeBased on video,this paper uses computer vision technology to manage the parking of vehicles,and studies the key problems in the video-based roadside parking management:vehicle tracking and license plate recognition.The main work of this paper are as follows(1)The vehicle tracking algorithm based on DeepSORT was studied to track passing vehicles in complex roadside environment and identify parking behavior through parking space occupancy ratio.In addition,the algorithm cooperates with the dynamic license plate recognition algorithm to recognize the license plate while tracking the vehicle,so as to prevent the license plate occlusion after the vehicle is parked and can not being recognized.Experiments show that the proposed algorithm can effectively identify the parking behavior of vehicles(2)The license plate recognition technology based on video is studiedAiming at tilting and twisted license plate,his paper analyzes the shortcomings of target detection algorithm in locating tilting and twisted license plates:the target detection algorithm can not accurately locate the plate,which affects the accuracy of subsequent recognition.In this paper,the text detection algorithm EAST is used to accurately locate the four vertices of the license plate,and then according to the results of the four vertices,the license plate is corrected to get the license plate in the front direction.Finally,the license plate is recognized by CRNN network.Experimental results show that the proposed method has a high accuracy of license plate recognition.The recognition rate of license plate in the front direction reaches 99.4%,and the recognition rate of license plate with large inclined angle is significantly improved.Compared with the RPNET proposed in the CCPD data set paper,the method is improved by 3.7 percentage points,reaching 96.2%As most of the roadside parking spaces are side parking spaces,the license plate will be blocked after the vehicle stops in the parking space.Aiming at the problem that the license plate is blocked,this paper studies the license plate recognition based on video,and designs the dynamic license plate recognition algorithm:When the vehicle enters the lens,the recognition begins.The license plate recognition results are obtained by the license plate recognition algorithm for each frame,and the license plate quality of the frame is scored based on the size of the license plate,the degree of license plate deformation and the degree of image blurring.Finally,the quality score is taken as the weight,and the best result is selected by voting algorithm.The dynamic license plate recognition method can solve the problem that the license plate is blocked after the vehicle is parked in the parking space,and make full use of the correlation between video frames to improve the license plate recognition rate.(3)The prototype of roadside parking management system is realized,and the effectiveness of the proposed algorithm is verified through experiments.In the experiment of license plate character recognition,a large number of license plates of various provinces are generated by the method of synthesis for training,and interference factors such as smoothing and tilting are added to simulate the real environment to solve the problem of fewer license plates of some provinces in the data set.