Study on Fusion of Terrestrial 3D Laser Point-clouds and Camera Image Data

Abstract

Some point-cloud holes usually exist in the point-cloud data of trees acquired by a terrestrial 3D laser scanner. Such holes can affect the integrity of point-cloud data and subsequent 3D reconstruction work. In order to solve the problems such as point-cloud holes or incomplete point-cloud data caused by the shield of obstacles or unable setup of 3D laser scanner, this study focuses on the registration of point-cloud data and repairing the point-cloud holes aided by photogrammetry after analyzing the point-cloud data, and discusses the precise registration of point-cloud data based on ICP (Iterative Closest Point) algorithm and the fusion of scanned point-clouds with point-clouds generated by images.The purpose of this study lies on the registration of the target point clouds acquired by a terrestrial laser scanner with the point clouds generated from the entity image data, so as to ensure the integrity of the entity point-cloud data.Firstly, a terrestrial 3D laser scanner is used to acquire the point-cloud data of two Chinese pine (Pinus tabulaeformis) trees in the scanning region, and at the same time the algorithm of ICP is used to register the scanned point clouds.Then, in order to make up the pointcloud data holes caused by external objective factors, a digital camera is used to take pictures of the Chinese pine trees and acquire the image data that has high degree of overlap.The pairwise matching of homonymy feature points of the adjacent images taken on site is completed by using SIFT (Scale Invariant Feature Transform) image stitching algorithm, and then PMVS (patch-based multi-view stereo) algorithm is used to generate the 3D point set of the two target Chinese pine trees.Finally, in the VC++ environment the ICP algorithm is used to fuse the point-cloud data obtained by 3D laser scanner with the 3D point set generated by PMVS algorithm.The results show that the mean square error for point-cloud registration and fusion are 0.0353733 and 0.0009226364, respectively, which indicate that the effects of the registration and fusion are satisfied.This research can accurately and quickly finish the registration of the point-cloud data obtained in different ways, solve the defects of point-cloud holes caused by objective factors in complex forest environments such as trees blocked or site settings and other factors, and realize the acquirement of complete three-dimensional point cloud of trees, which has played a key role for subsequent three-dimensional reconstruction and parameters extraction of the trees.

Code and data availability

The full-text asset investigation has not produced a conclusive result yet.

No evidence-backed public reproduction asset is currently recorded.