Research And Implementation On The Method Of Automatic Ally Creating Panoramic Map With UAVs

Panoramic maps combine panoramic image information with digital maps to provide people with new-type map services,which has been widely used in recent years.However,the creation of panoramic maps usually needs excessive labor and time costs.Utilizing a Unmanned Aerial Vehicle(UAV)platform equipped with camera and achieving the whole process of building a panoramic map automatically can save labor and time resource and improve efficiency.It is a novel intelligent solution for creating panoramic maps,which has important theoretical research value and practical application prospects.In order to realize the autonomous generation of panoramic maps with UAV platform,two core problems need to be addressed:autonomous generation and optimization of waypoints and optimization of shooting positions for UAVs.The former aims to answer the question of where the UAVs should shoot.To this end,this thesis proposes an effective shooting position selection rule,designs autonomous waypoint generation and optimization algorithms and achieves the goal that the task of manually selecting waypoints is handed over to computer to generate and optimize the waypoint quantity to improve the efficiency and save labor.The latter aims to fine-tune UAV's location and height to ensure the road is in the center of image and the whole building is captured,so that we can better achieve the image stitching and improve the overall display effect of panoramic maps.This thesis proposes the Road Following Transfer Net(RFTNet)to make UAV follow the center of road and an sky area detection algorithm to indirectly adjust the height of UAV dynamically when photographing buildings;Finally,a prototype system is implemented in this thesis,and multiple datasets are established in a real campus or a virtual world.Extensive experiments and performance comparisons between our algorithms and other baselines are fully analyzed.The results show our algorithms achieve better performance with less computational overhead.