Design Of Smart Parking Lock Based On Machine Vision

Recently,with the increasing number of private cars,most cities face a shortage of parking space.In some towns,the number of parking lots is only 1/3 of that of the vehicles.In such a situation,parking difficulty has become progressively significant.The resulting phenomenon of parking chaos and looting of parking spaces is widespread.Therefore,there emerge many parking locks for efficient parking management.At present,the parking locks on the market are either mechanical or remotely controlled.As a result,there still exist some problems,such as timeconsuming and labor-consuming manual unlocking,the short distance of remote control,and a low degree of automation.On the other hand,with the perfection of license plate recognition technology and the continuous update of algorithms,vehicle management system based on this technology is widely used on many occasions,such as parking lot,road monitoring.This thesis thus applies the license plate recognition technology to the parking lock to realize automatic unlocking and locking.However,the high-precision license plate recognition system based on a deep neural network model often needs to occupy enormous equipment resources.It is challenging to run in low-cost equipment such as parking locks.In addition,the algorithm based on the conventional image-processing method usually has low accuracy and high pixel resolution requirements.In order to solve the above problems,based on the research of license plate location and character recognition algorithm,this thesis establishes an efficient license plate recognition model suitable for embedded hardware and for balancing resources and performance.Combining the model with the parking lock device,a smart lock device based on machine vision is designed.Based on the above,the main part of the thesis can be summarized into the following three aspects:(1)The image segmentation network is applied to the license plate location algorithm,and a license plate location model based on the medical image segmentation U-net network is designed.Because of its lightweight structure,its adaptation to any size of images,and its good segmentation effect,the U-net network is suitable for the license plate location model.Therefore,this thesis trains a high-performance location model based on the U-net network with the preprocessed small sample data set.Compared with the image-processing positioning algorithm,the results show that the license plate location accuracy based on the U-net network reaches 96% in complex scenes,much higher than another algorithm.(2)Based on the systematic analysis of the license plate character recognition algorithm,the gray wolf optimization(GWO)algorithm is adopted in character recognition model based on the support vector machine(SVM)and a GWO-SVM domestic license plate character cascade recognition model is constructed.The GWO algorithm optimizes the penalty factor and kernel parameters,and other mainstream SVM-based optimization models are compared and analyzed.The accuracy of the character recognition model has been improved by more than 5% on average,and the algorithm has fast iteration and high character recognition accuracy.In addition,the Unet positioning model is combined with the GWO-SVM character recognition model to establish a new type of embedded license plate recognition algorithm UNET-GWOSVM model with high resource utilization.The test results show that in the embedded hardware environment,the recognition accuracy of the algorithm can reach 94%,and the average time of recognition is controlled within 2 seconds.(3)Combined with the above license plate recognition model,a smart ground parking lock based on machine vision is designed.The parking lock uses the Raspberry Pi development board Raspberry＿Pi3B as the core controller.It uses the ultrasonic ranging module to sense the vehicle and the camera to collect license plate information.Then the UNET-GWO-SVM model is used to recognize the license plate.It is compared with the bound license plate number to determine the opening of the bracket lock.In addition,the device also establishes a connection with the Internet of Things platform through the 4G transmission module,which is convenient for users to manage the lock through the mobile terminal and realize parking space sharing.To verify the performance of the car lock,a field test is conducted on campus.The test results show that when the license plate matching similarity is 85%,the effective passing rate is98.04%.When the license plate matching similarity is 70%,the effective passing rate reaches 99.02%,satisfying the design requirement.