Page 647 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

LoD0 is display of building represented by footprints or roof-edge polygons. LoD1 is block with flat roofs and vertical
walls. In LoD2, there are distinct boundaries between different structures of roofs and surfaces with installation of balconies and
dormers. LoD3 includes the most detailed architectural exterior such as doors and windows. LoD4 has the exterior appearances
same as LoD3 models, but interior structures like rooms, furniture are added [12]. So, degree of closeness and abstract between
model and real-world are indicated by LoD. Extracting information about buildings is one of important steps in LoD generation.
In this article, LoD is created and an approach is presented for LoD generation.
Recently, several approaches have been proposed for generating 3D building models. In [13] an approach is proposed
that integrates the various levels of data. A hierarchical representation of 3D building models is used to combine different data
sources including aerial and ground view images. Using this method makes generating building models feasible. The proposed
system has ability to construct complex building more than other image based automatic systems. In the other work [10] to
generate a complete model of building in high Level of Details, combination of TLS and ALS data is necessary. Wavelet-based
is used to process and combine data from ALS and TLS. In proposed approach, methods of selecting tie points are applied to
integrate point clouds in different datum.
In [14] presented a method to reconstruct building model using LiDAR data. In this method building contours are
reconstructed using a graph-based approach. Finally, building models are derived by analysis of building contours based on
hierarchical structure. First for illustration the topological structure of buildings, a graph theory-based is applied to localize
contour tree method. Then, analyzing relationships between topological structure of buildings causes to separate the buildings
into different parts. Finally, combination all models derived through the bipartite process makes reconstruct the building model.
Proposed method can reconstruct complex buildings models with mean modeling error of 0.32. [15] presented an algorithm to
construct 3D models of building in urban environment using ALS data. Irregular point clouds are applied to extract planer faces
by using 3D Hough transform. There are two different strategies to reconstruct building models by using planer faces and
segmented ground plans. In the first strategy, lines and height jump edges are recognized and intersected together. In the other
strategy all planar faces are assumed some part of the building. Whereas second strategy has ability to reconstruct more building
and more details of this building but reconstruct parts of model that not exist.
In [16] a data-driven method is proposed to reconstruct building from LiDAR point clouds. In this approach, first a 2D-
grid is covered the segmented point cloud. Second, in every grid cell 3D vertices of building model can match with the
corresponding LiDAR points. Then, quad-tree method reduces the number of 3D vertices, and for connection the remaining
vertices use their nearness in the grid. Triangular Irregular Network (TIN) illustrates roof segments and applying common
vertices or – at height discrepancies – vertical walls to connect them to each other. 3D building models derived from this approach
have a very high accuracy and level of detail, due to containing roof superstructures like dormers.
In this work, an approach is presented for 3D building modeling. In our work, both of ALS and MLS point clouds are
used. Roofs and facades of building are extracted. Walls are seperated based on Ransac algorithm.

2. Methodology
The workflow of the proposed algorithm is shown in Figure 2. The algorithm involves two main steps:
1) Pre-Processing: this step includes sectioning to reduce data volume and speed up implementation, noise removal and
remove points of ground.
2) Building extraction and modeling: first, roofs are separated and next facades are segmented. At the end of that, building
utility modeled.

.
Figure 2. The flowchart of the proposed algorithm
2