We congratulate Jules MOREL, PhD candidate at the IFP / LSIS, Aix-Marseille University, France, for having successfully defended on February 17, 2017, his PhD thesis entitled “Surface Reconstruction Based on Forest Terrestrial LiDAR Data”, the abstract of which is provided below:
Abstract:
In recent years, the capacity of LiDAR technology to capture detailed information about forests structure has attracted increasing attention in the field of forest science. In particular, the terrestrial LiDAR arises as a promising tool to retrieve geometrical characteristics of trees at a millimeter level.
This thesis studies the surface reconstruction problem from scattered and unorganized point clouds, captured in forested environment by a terrestrial LiDAR. We propose a sequence of algorithms dedicated to the reconstruction of forests plot attributes model: the ground and the woody structure of trees (i.e. the trunk and the main branches). In practice, our approaches model the surface with implicit function build with radial basis functions to manage the homogeneity and handle the noise of the sample data points.
Our first focus is on the reconstruction of the ground surface whose level of detail is based on local complexity, through alternation between scale refinement, filtering and reconstruction. The result arises from the polygonization of the implicit function expressed as the merging of local approximations by compactly supported radial basis function used as partition of unity. Once the ground is modeled, the topology effects can be ignored in the following computation steps that focus on the modeling of trees. Traditionally, the processing of the woody part is achieved by a discrete reconstruction in the form of a stack of independent building blocks. From such a model, our approach developed for the ground is adapted to approximate the woody part of the tree by a more flexible continuous surface. Expressed as an implicit function, the tree model can be refined by an additional computational step in order to describe precisely the geometry. With this in mind, we propose a method dedicated to the fine reconstruction of occluded objects: from 3D samples presenting occlusions, we use the previously described continuous model to guide a Poisson surface reconstruction. Thus, we guarantee the production of a watertight surface that approximates sharply the point cloud in the visible areas and extrapolates consistently the tree shape in the occlusions.
Abstract:
In recent years, the capacity of LiDAR technology to capture detailed information about forests structure has attracted increasing attention in the field of forest science. In particular, the terrestrial LiDAR arises as a promising tool to retrieve geometrical characteristics of trees at a millimeter level.
This thesis studies the surface reconstruction problem from scattered and unorganized point clouds, captured in forested environment by a terrestrial LiDAR. We propose a sequence of algorithms dedicated to the reconstruction of forests plot attributes model: the ground and the woody structure of trees (i.e. the trunk and the main branches). In practice, our approaches model the surface with implicit function build with radial basis functions to manage the homogeneity and handle the noise of the sample data points.
Our first focus is on the reconstruction of the ground surface whose level of detail is based on local complexity, through alternation between scale refinement, filtering and reconstruction. The result arises from the polygonization of the implicit function expressed as the merging of local approximations by compactly supported radial basis function used as partition of unity. Once the ground is modeled, the topology effects can be ignored in the following computation steps that focus on the modeling of trees. Traditionally, the processing of the woody part is achieved by a discrete reconstruction in the form of a stack of independent building blocks. From such a model, our approach developed for the ground is adapted to approximate the woody part of the tree by a more flexible continuous surface. Expressed as an implicit function, the tree model can be refined by an additional computational step in order to describe precisely the geometry. With this in mind, we propose a method dedicated to the fine reconstruction of occluded objects: from 3D samples presenting occlusions, we use the previously described continuous model to guide a Poisson surface reconstruction. Thus, we guarantee the production of a watertight surface that approximates sharply the point cloud in the visible areas and extrapolates consistently the tree shape in the occlusions.