| dc.description.abstract | Car driving in the parking garage to find a parking space without any guidance mechanism often takes a lot of time. Maybe need to worry about taking a long detour to find a parking space, which is a long and laborious task, so this paper aims to establish an indoor parking system guided by a drone. First, the parking garage map needs to be built into the system. Then the system automatically screens the parking space closest to the parking garage entrance, plans the shortest path to this parking space, and then guides the car according to this path through the drone. After the guidance mission, the drone will take an image of the car that has completed the parking and send it back to the system to update the parking space status. Such a system can save drivers′ parking time and make parking lot management more effective.
This guidance system consists of two subsystems: the parking garage and drone systems. The parking garage system includes parking space selection and path planning. The system will evaluate each vacant parking space through the A* algorithm to find the target parking space closest to the entrance and plan the shortest path for the drone guidance. The drone system includes indoor positioning and guided car status tracking method. Because the GPS signal cannot be received indoors, the system is based on the AprilTag to assist the drone in completing the flight guidance task. The guided car status tracking method can be divided into four functions. The first is car detection, and the Yolov4 tiny, lightweight network model is used to do car detection, which is sufficient to achieve the detection effect and reduce the computing resources of the embedded system. The second is car tracking, by improving the SORT algorithm(Simple Online And Realtime Tracking) to track guided car dynamics. The third is the search mechanism, and the system will automatically start the function of searching for the tracking target when the tracking fails, find the tracking target feature through the ORB algorithm(Oriented FAST and Rotated BRIEF), and then use the FLANN matcher(Fast Libary for Approximate Nearest Neighbors) to do target pairing, the target will be successfully retrieved finally. The fourth is the waiting mechanism, the drone can keep an eye on the distance between the guided car and the drone at any time during guidance, and if the distance is too far, the drone will stop and wait for the car to follow up to implement the guidance purpose. This paper verifies the system′s feasibility in the actual parking garage environment. The measured results show that the drone can achieve the required tasks in a non-interference environment. | en_US |