Research And Implementation Of Key Technologies Of UAV-based Navigation

Due to the complexity of the UAV flight environment,the traditional vision-based navigation system and GPS signal-based navigation system will be affected by light conditions and GPS signals,so that the UAV may have the problem of positioning failure.The autonomous navigation system of radar has strong robustness and can better cope with the complex flight environment of UAVs.The research content of this thesis is that UAVs only use lidar to perceive the environment without relying on visual sensors and GPS signals,focusing on the generation and denoising of high definition map,point cloud ground segmentation and UAV positioning.The research content is carried out from the following aspects:（1）According to the functional requirements of the Li DAR autonomous navigation system,this thesis presents a design scheme of the Li DAR-based UAV autonomous navigation system,which divides the entire system into the UAV side and the ground station side,including high precision.Map generation module,three-dimensional NDT positioning module,autonomous obstacle avoidance navigation module,etc.,the software structure and hardware selection of the system are given,and the software and hardware are combined to complete the whole system function.（2）For the problem of high definition map generation and denoising,this thesis generates the high definition map used in the system based on the LEGO-LOAM algorithm.In view of the shortcomings of the loopback detection part of the algorithm,the Scan Context global descriptor is used for optimization and improvement.The EVO assessment tool quantitatively analyzes the effectiveness of the improvement.For the noise existing in high definition map,a DBSCAN algorithm based on voxel division is proposed,which divides point cloud noise into large-scale noise and small-scale noise,uses neighborhood search combined with statistical characteristics to remove large-scale noise,and uses voxels to remove large-scale noise.Lattice division combined with density clustering algorithm to deal with small-scale noise to achieve optimal denoising effect.（3）Aiming at the problem of point cloud ground segmentation and UAV positioning,this thesis proposes a ground segmentation algorithm based on area division.Combined with the scanning characteristics of lidar,the 3D point cloud data is divided into multiple areas,and the analysis is carried out in each sub-area.For plane model fitting,in order to optimize the accuracy of plane model fitting,raster projection combined with elevation information is used for rough screening.According to the complexity of each sub-region,adaptive threshold is used as a segmentation index for dynamic segmentation.Then,the ground segmentation algorithm based on area division was applied to the 3D NDT algorithm,and the experimental comparison was carried out from the perspectives of registration time and registration accuracy,and the impact of ground information on UAV positioning was quantitatively analyzed.