| Slow parking in big cities brings a lot of inconvenience to people’s lives.Many parking lots now need to lift poles to enter,and a lot of manpower and material resources need to be invested in the construction and planning of the parking lot,as well as the subsequent maintenance,which requires a lot of costs.At present,artificial intelligence and deep learning have many landing projects in different fields,and have achieved very good expected results.Combining artificial intelligence and deep learning algorithms to create a completely unsupervised parking lot,there is no need to build many building facilities,only cameras are needed to monitor the parking lot,and the camera can detect parking spaces,identify license plates,and timekeeping functions to form a set Completed process system.Compared with the currently widely used parking lot supervision method,it can save a lot of financial and material resources.There are two types of parking space detection in general.One is to identify the parking space based on the line identification of the parking space.The second is to detect parking spaces based on the image information in the parking spaces.Both methods have advantages and disadvantages.Judging from the parking line,the disadvantage is that the parking line is blocked during the recognition process,or the changing factors of the environment will fade after a long time of use and it is not easy to distinguish.The disadvantage of the detection of vacant parking spaces is that the judgment of the vacant size is inaccurate.The detection of parking spaces according to the image information in the parking spaces requires real-time tracking of the vehicle,which requires a higher computing power for the computer.The mainstream method of license plate detection is mainly based on Open-cv recognition.The application scenarios are all parking lot lifting poles,and the vehicle enters from a predetermined position.This requires the angle of the license plate and the high definition to meet the conditions to recognize the license plate information.The main research content of this article focuses on the detection of parking spaces with and without standard lines.How to use the improved gradient feature extraction to detect the occupancy of parking spaces through the method of pixel calibration.Using the improved texture feature extraction method,the improved step-by-step corner detection method positioning,and the improved network framework,how to identify and detect in the long-distance,unclear license plate,and multiple license plate images,to achieve the purpose of accurate and rapid recognition.The main research contents are as follows:(1)In terms of parking space detection,this paper proposes a diamond calibration method,which can reduce the interference of parking spaces next to it and reduce the range of the gradient acquisition area,achieving high recognition rate and real-time detection requirements.The difference frame strategy is adopted,and the improved HOG(Histogram of Oriented Gradient)algorithm is used to collect the gradient changes before and after the time of each parking space at a fixed time period,and compare the difference of the gradient change range before and after,so as to achieve the detection of parking spaces.purpose.(2)In terms of vehicle license plate detection,first use the improved LBP(Local Binary Pattern)algorithm combined with the Adaboost algorithm for rough positioning.Then use multiple thresholds to determine the ROI(region of interest)area for improved step-by-step corner detection,and finally fit the corner edge features to determine the precise location of the license plate.(3)Pass the obtained license plate image into the improved lightweight CNN(Convolutional Neural Networks)framework,and improve the convolution method,the type of activation function,and the method of pooling.In this way,the FPS of the image can be greatly improved to reach the level of smooth recognition when the recognition rate has been improved to a certain extent. |
PDF Full Text Request