Research On Key Technologies Of UAV Indoor Positioning And Mapping

Multi-rotor UAVs have the advantages of small size,strong maneuverability,low cost,and easy operation,which make them have great advantages in the fields of search,rescue,and military reconnaissance.For multi-rotor UAVs,robust and high-precision positioning and mapping technology is the key to achieving autonomous indoor and outdoor flight.In recent years,positioning systems based on Inertial Navigation System(INS)and Global Navigation Satellite System(GNSS)have been greatly developed.However,in some complex environments such as indoors and factories,due to problems such as multipath signal interference and intermittent rejection,the GNSS-based positioning system is still prone to errors,and the INS positioning solution will accumulate errors rapidly,resulting in unmanned equipment.Incorrect location information.VisualInertial-Odometry(VIO)does not depend on GNSS and has the advantage of local high-precision positioning.However,the VIO system has the problem of easy loss of positioning and inability to perform global positioning in scenarios such as illumination changes and missing features.To meet the needs of seamless and high-precision positioning in GNSS-rejected environments such as indoors and factories,this paper is based on vision,GNSS,and Inertial Measurement Unit(IMU)sensors.Visual Simultaneous Positioning and Mapping(Visual Simultaneous)Based on Localization and Mapping(V-SLAM)technology,the indoor and outdoor seamless high-precision positioning method based on visionGNSS-IMU fusion is deeply analyzed and studied.Secondly,while achieving robust and high-precision positioning in a GNSS intermittent rejection environment,to meet the needs of UAVs to explore unknown environments,this paper presents a high-precision 3D reconstruction technology for indoor scenes based on consumer-grade RGB-D cameras.In-depth analysis and research.This paper mainly completes the following aspects of the work:1.Based on the Microsoft AirSim platform and based on the UE4 simulation engine,a UAV simulation environment including factories,cities,and wild forest scenes has been built,and a set of sensors such as GNSS,IMU,and vision has been developed for this simulation environment in real-time.Data application interface.Secondly,with the PX4 open-source flight control as the core,using GNSS as the outdoor positioning solution,and using VIO as the indoor positioning solution,a quadrotor UAV platform capable of stable indoor and outdoor positioning and navigation was built from scratch.2.Aiming at the problems that the UAV based on the VIO positioning system is easy to lose visual positioning when flying in a low-textured environment,and the GNSS positioning intermittently refuses to cause the UAV to fail to work properly,this paper proposes a method based on nonlinear optimization to achieve The fusion of GNSS and VIO positioning information,GNSS assists VIO to improve the robustness of the system in a good signal environment,and uses VIO to obtain global high-precision positioning in a GNSS-denied environment,and finally realizes indoor,factory,etc.A robust and consistent positioning solution for UAVs in GNSS intermittent denial environments,compared to differential GNSS positioning errors of 1-3 meters,the root mean square error of the system’s long-term flight trajectory can be kept within 1 meter.3.Aiming at the problems of large volume,missing details,less texture,and too much noise in traditional dense point cloud reconstruction schemes,this paper proposes an indoor high-precision and lightweight reconstruction method based on consumer-grade RGB-D cameras.After the dense point cloud map is obtained based on the VIO pose and RGB-D data,it is simplified to a grid map.At the same time,for the problem of VIO pose errors,a global optimization system based on luminosity consistency is used for VIO.The pose is optimized,and the image texture is projected to the grid map based on the optimized pose,and finally,a lightweight,texture-rich,and high-precision indoor reconstruction result is achieved.Compared with the original dense point cloud map,the memory usage efficiency is increased by an average of 68.34%.