Three-dimensional Reconstruction Of Low-rise Building Scenes Based On Airborne LiDAR

With the development of computer computing power and hardware equipment,in order to further improve the accuracy of detection target parameters.Researchers have found that airborne lidar has better detection performance than traditional millimeterwave radar,and its three-dimensional reconstruction characteristics have become a hot spot in the field of development.In addition,in order to promote the application and development of airborne lidar in actual scenarios.Thesis studies the 3D reconstruction of low-rise building scenes with airborne lidar.The specific research content is as follows:When the point cloud data of low-rise building scenes is obtained for Li DAR,due to the large area of low-rise building scenes,the large amount of point cloud data,and the deviation of UAV flight routes,the data has the problems of uneven distribution,lack of significant features,redundancy and absence.An improved outlier point denoising algorithm based on statistical features is proposed,and the index structure of point cloud data is improved.The simulation results show that the algorithm can effectively reduce the data volume and improve the data accuracy.Aiming at the problems of point cloud voids on the scene surface and incomplete point cloud data model caused by scanning blind zone when airborne lidar reconstructs the 3D model of low-rise building scene.An improved point cloud registration algorithm for low-rise building scenes based on convolutional neural network is proposed.The algorithm uses the RANSAC algorithm to eliminate mismatched point pairs and find the correct feature point pairs.Combined with the feature matching method based on Euclidean distance,the point cloud coarse registration is performed,and then the improved ICP algorithm is used to re-register the point cloud.The simulation results show that the algorithm effectively solves the problems of point cloud voids and incomplete cloud data,and the three-dimensional low-rise building scene reconstructed by the algorithm is more accurate.In view of the high frequency of airborne lidar data collection,the large amount of point cloud data and the lack of overlapping scenes required for registration between point cloud data.A reconstruction strategy of block fusion is proposed.Compared with the original method,this strategy achieves high-precision fusion of point clouds.Moreover,this strategy uses the abnormal matching detection based on the Mahalanobis distance and the transformation matrix of track fitting to adjust the outliers,uses a sliding window to intercept the global point cloud,and matches the current point cloud block with the intercepted sub-point cloud respectively.Quasi-calculation to determine the transformation relationship between each point cloud block and the global coordinates,and realize the coordinate transformation of all point cloud blocks.Simulation results show that this strategy effectively solves the problem of point cloud data overlap,and this strategy can perform point cloud registration between adjacent frames,thereby realizing reconstruction of low-rise building scenes.