Research On Unmanned Parking Lot Using CNN Technology

In recent years,the car ownership has increased year by year,while the supply of parking spaces is insufficient and the utilization rate is low.The contradiction of parking difficulty in many cities has become increasingly prominent.Aiming at this problem,the design of unmanned parking lot is studied in order to improve the utilization and throughput of existing parking lot and change parking difficulty.The designed unmanned parking lot has three functions: parking space state detection,parking space guidance and reverse car search.The function of parking detection is to collect,identify,update and upload parking information through the parking recognition terminal composed of Raspberry Pie and camera;The function of parking guidance is to use Floyd algorithm to calculate the shortest path between the user and the target parking space,and mark the path on the map;Reverse car search is a Wi Fi-based fingerprint location method to help users find cars.Finally,the system of interaction between server and mobile client is designed.Users can view the recommended parking space or custom parking space,path guidance schematic diagram and their own positioning schematic diagram through mobile phone.In the design process,the low accuracy of parking space state detection and the poor performance of clustering boundary location in fingerprint location algorithm are improved.A mini-convolutional neural network Mini Mobile Net is designed for parking space state detection,which can run on the Raspberry Pie.This network has a small size and a small number of parameters.In the same test set,the accuracy rate reaches more than 99%,and in different test sets,the accuracy rate reaches95%;Aiming at the problem of poor positioning performance near the class boundary caused by clustering fingerprint database in reverse search,the idea of minimum membership interval is proposed.The fingerprint database is managed by cross-soft partition,so that fingerprints that can not be clearly classified can be divided into multiple sub-fingerprint databases,the average positioning error near the class boundary is reduced by 0.72 meters.At the same time,the fuzzy C-means is vulnerable to the initial value,and the clustering results are unstable.The improved particle swarm optimization algorithm is used to solve the clustering center instead of the iteration method.The improved particleswarm optimization algorithm combines satisfaction with linear decreasing inertia coefficient and introduces mutation.The optimization ability is better,and the fuzzy C-means clustering results fused with the improved particle swarm optimization algorithm are stable and not affected by the initial value.Finally,the function test of the unmanned parking lot system is carried out.The test results show that the parking lot identification terminal can complete three functions: the capture and identification upload of the parking space.The mobile client can query and display the parking space,display the path guidance schematic map and reverse car search.