Research On 3D Scene Reconstruction Based On Airborne LiDAR

LiDAR(Light Detection and Ranging)is a remote sensing technology that has a rapid development in recent years.By actively transmitting and receiving laser pulse signals,it obtains spatial information of targets at a high speed,which provides efficient and accurate 3D data for environmental modeling,so it has been widely used in 3D Scene Reconstruction.The3 D scene reconstruction technology realizes digital representation and processing of real scene information in computer,becoming a key issue for many research and applications in the fields like computer vision,virtual reality and autonomous driving.Airborne LiDAR fixes LiDAR on the UAV platform perform to realize high-altitude scanning of outdoor scenes and restore real spatial information of the ground and ground objects.This paper studies 3D scene reconstruction technology based on airborne LiDAR,including laser data acquisition and processing,point cloud registration algorithm and reconstruction strategy,and carries out corresponding optimization and improvement according to characteristics and practical applications of airborne LiDAR data.This article first builds a complete airborne LiDAR data acquisition platform,which integrates a UAV with flight controlling system,a multi-scan LiDAR and an embedded computer,to collect and process actual outdoor scene data.Data acquired by the system is a series of discrete and irregularly distributed 3D coordinate points,which is called "point cloud".In this paper,the outlier removal method based on the statistical filter and the down-sampling method based on the voxel filter are used to pre-process point cloud data to improve the quality and efficiency of reconstruction.Aiming at the problems of uneven resolution,sparse irregularity,low overlap rate and numerous redundant ground points in airborne LiDAR point cloud data,this paper proposes an intensity-constrained registration algorithm with an initial estimation algorithm based on ground segmentation combining motion information.First,the rotation transformation relationship of adjacent frames can be calculated by the normal vector of ground point clouds segmented by an improved RANSAC(Random Sample Consensus)algorithm.Combining the rotation parameters with the translation information obtained by UAV platform provides good initial value estimation for registration,on the basis of which an intensity-constrained Generalized-ICP registration algorithm is proposed.This method makes full use of the intensity information of LiDAR to constrain the corresponding point set participating in the iteration to achieve the precise registration.The experimental results show that the proposed algorithm is able to obtain the accurate transformation relationship between laser point cloud data and has higher registration accuracy and efficiency.Finally,this paper proposes a reconstruction strategy of group fusion.First,the original frames are registered and merged into groups of point cloud blocks with relatively uniform resolution,which overcomes the impact of uneven resolution on the global registration.In this process this paper proposes to use Mahalanobis distance to detect abnormal matching and proposes a correction method based on track fitting to adjust mismatched matrixes.For the global fusion of formed blocks,a sliding window method is proposed to obtain direct relationships from blocks to the global coordinate system.Frame fusions and block fusions are executed alternately and serially until the complete reconstruction result is achieved.Through experiments,the implementation process of the algorithm and the reconstruction results in different scenarios are shown.To sum up,this paper builds a simple,low-cost airborne LiDAR system,carries out practical experiments to research and optimize the key technologies of 3D scene reconstruction,making improvement in accuracy and efficiency.Real and effective restoration and visualization of spatial information of those complex outdoor scenes are accomplished with good resolution and accuracy.