Page 12 - 中国仿真学会通讯2020第1期
P. 12

KEYWORDS
Building point cloud, Three⁃dimensional modeling, Hierarchical reconstruction, Block
reconstruction

1  Introduction                                      to extract the roof edge, and stretches the roof
                                                     edge to build the 3D model of buildings. All the
Photorealistic three⁃dimensional ( 3D ) model of     methods mentioned above aim at simpler models
buildings is the foundation of digital cities, and   and only consider the geometric information of the
has potential application prospects in areas such    building roof. Then, Wan [ 7 ] proposes a
as urban planning, urban management,automatic        grammatical rule⁃based method for the
driving and other fields, which have attracted       reconstruction of 3D building wall. By introducing
great attention from the academic community [1] .    grammatical rules, the building is reconstructed
With the continuous development of the computer      into a quadrilateral set. Wang [ 8 ] proposes a
technology and the mapping technology, such as       modeling method based on line frame. In order to
the airborne laser scanning system, the vehicle      establish the wireframe model of 3D building, the
laser scanning system, the aerial photography of     extracted wireframe is optimized by the generative
drones, and the aerial oblique photogrammetry,       adversarial nets with conditional constraints. The
the 3D reconstruction of buildings has been          methods mentioned above provide relatively
greatly improved in terms of automation and          complete point cloud of the building facade,
modeling effects.                                    which can improve the effect of the building
                                                     facade model. However, such methods do not
      In the traditional way, we can obtain a large  consider the effect of building roof modeling, and
number of building point clouds through laser        the modeling process is affected by the factors
radar, including airborne and vehicular, and         such as the LIDAR acquisition equipment, the
obtain a complete building point cloud model         acquisition condition and model occlusion. All
through data fusion. The traditional modeling        these factors need a certain degree of manual
method uses the two⁃dimensional ( 2D) vector of      interaction, which result in low automation. In
a building to create 3D white model of the           addition, Zhang [ 9 ] proposed a building
building, and the texture is obtained by manual      reconstruction method based on structural
shooting [ 2] . This method is low in automation     perception, which can solve the problem of
and mode laccuracy. The development of LIDAR         missing point cloud on the building surface.
technology enables scholars to obtain a large
number of 3D point cloud of buildings. At the              At present, aerial oblique photogrammetry
same time, a series of methods for 3D                refers to acquiring an image using an aerial
reconstruction of buildings based on LIDAR point     camera with a certain oblique angle. It is mainly
cloud have been proposed. Lafarge [ 3, 4 ]           equipped with multiple sensors on the same
regards the buildings as a collection of simple      information acquisition platform, which can
urban structures, similar to Lego bricks. They       collect more complete and accurate information of
extract 2D models from the roof point cloud and      ground objects from different angles such as
build the model by placing blocks of 3D models       vertical and oblique. The rapid development of
on the 2D models. Poullis [5, 6] uses the energy     aerial oblique photogrammetry provides a large
minimization method based on graph cut algorithm     amount of raw data for 3D reconstruction of

                                                                                                             09
   7   8   9   10   11   12   13   14   15   16   17