Research On Registration Techniques Of Building's LiDAR Point Cloud From Different Views

Laser scanning technology is characteristic with continuous,automatic,not contact,and fast acquisition in high resolution space of 3D data,which is gradually becoming a research hotspot that contain three-dimensional building reconstruction and digital city.In order to build a complete building's LiDAR point cloud data,we need to registration the multi-view building's LiDAR point cloud data.In this paper,we take the 3D building reconstruction as the background,mainly studies the problem of registration different view building's LiDAR point cloud.The main work is as follows:1.Aiming at the problem that the registration of building's LiDAR point cloud lack of precision,which registrate the point cloud based on FPFH feature descriptor and eliminate error matching points only based on setting distance threshold,we put forward a kind of improved registration algorithm.On the basis of the eliminating the error matching points by distance threshold,we add the constraints of the eliminating duplicate and the normal vector of the key points to eliminate the rest of the error matching point;finally,using the rest of the matching points to complete the registration of the building's LiDAR point cloud.Experiments show that this improved algorithm can complete the registration of the building's LiDAR point cloud,and improve the accuracy of registration.2.In view of the traditional CPD registration algorithm with high computational complexity,the mass point cloud data is difficult to directly using the algorithm to registrate,in order to solve this problem,we put forward a kind of fast CPD building's LiDAR point cloud registration algorithm based on ISS key points.Firstly,for reducing the scale of the building's LiDAR point cloud data,using the ISS algorithm to extract the key points of the building's LiDAR point cloud;in the second place,using the CPD algorithm to registrate the ISS key points which extracted from different view building's LiDAR point cloud.The experimental results show that the improved registration algorithm is simple and effective,stable and reliable,and improve the efficiency of building's LiDAR point cloud registration using the algorithm of CPD.3.Inspired by the registration algorithm of two-dimensional image and the rich structure information of buildings,we propose a building's LiDAR point cloud registration algorithm based on dimension reduction.The method adopt the idea of dimension reduction,firstly,in order to construct the building's projection plane,we use the least square method for fitting the facade of the building;secondly,for maximum limiting retains the inherent structure information of the building,we put the building point cloud vertical projection to the projection of the building;in the next place,resampling the projection points to generate two-dimensional image,and then using the template matching method to extract the corresponding points;finally,index the conjugate points of images back to the building's LiDAR point cloud data,then we can complete the registration of building's LiDAR point cloud.The experimental results show that the proposed registration algorithm can accurately obtain the corresponding points of building point cloud,and can effective to complete the registration of building point cloud.