Research And Design Of Fine Parking Management System Based On Machine Vision

With the rise of our national economy,people’s life is increasingly affluent,and the total number of private cars is also increasing.It is more and more difficult for people to park.Improving parking efficiency and improving the utilization rate of parking lot is the fastest and cheapest way to solve the current problem.Based on this,a set of fine parking management system are designed in this paper.Through machine vision technology,the free parking space is detected in real time and the vehicle model is identified.Through different vehicle sizes,the free parking space more suitable for its size is allocated to the vehicle.So that the manager can be more reasonable planning parking lot,more efficient use of parking space.Therefore,the accuracy and running rate of the parking space state detection algorithm and the vehicle type recognition algorithm are put forward higher requirements.Based on this,the main work of this thesis are as follows:(1)Aiming at the problem of insufficient computing power of embedded devices,a lightweight parking status detection model based on YOLOv5 s was proposed.Firstly,in order to improve the accuracy of the recognition of the YOLOv5 s model,the Convolutional Block Attention Module(CBAM),which has a strong ability in extracting important features,is used.The superiority of the optimized model is demonstrated by ablation experiments.Then,Network Slimming algorithm is used to carry out pruning operation on the BN layer of the optimized model.Lastly,by comparing and analyzing the detection effect of the models with different pruning rates,the optimal lightweight model is obtained.(2)In the study of vehicle type identification,an optimization algorithm based on Vision Transformer(Vi T)model is proposed.In order to extract image features better and reduce the amount of network computation,firstly,the original image is input into the Vi T model after three CBS convolution operations.Secondly,replace the MLP head module in the Vi T model with Conv MLP.Finally,through the ablation experiment and the comparison experiment with the detection results of the traditional CNN model,it is concluded that the identification accuracy of the optimized model is higher,which proves the effectiveness of the optimization scheme.(3)In order to reduce the data throughput and improve data transmission speed of the system,the overall system structure of collaboration between the edge,cloud and terminal is designed.The edge adopts the distributed structure of embedded devices.The embedded devices carry out preliminary detection and recognition of the video incoming from the camera,and only transmit the recognition results to the cloud.The cloud is responsible for the connection,processing and storage of incoming data from the edge,as well as the transmission of data to the terminal.In order to facilitate parking management and car owners parking,the terminal uses PC terminal,mobile terminal and display module to interact with users.(4)Design and implement fine parking management system.The overall functional structure module of the system terminal is designed,and the process of each functional module and the database used by the system are designed in detail.Finally,Win Form and Android were used to develop the software system,which realized user registration and login,parking lot query,parking space recommendation,Online payment,historical order query and other functions.