AN INTEGRATED SPATIAL INDEXING
OF HUGE POINT IMAGE MODEL
Yanmin Wang“, Ming Guo ^*
* Key Laboratory for Urban Geomatics of National Administration of Surveying, Mapping and Geoinformation, Beijing
University of Civil Engineering and Architecture, Beijing 100044, China - wangyanmin@bucea.edu.cn;
guoming@bucea.edu.cn
Commission IV, WG IV/4
ABSTRACT:
In order to manage point image model easier and build the point cloud model on the basis of the need and improve the query
efficiency, a 2D&3D integrated spatial index of large point image model is proposed in this paper. Point image is managed by 2D
quad tree and 3D MBB. Point image model constituted by point image are indexed by 3D-R tree. Finally the organized hierarchical
model and other attribute data are stored in commercial database. The storage, management and visualization of large point image
model are verified on the ordinary PCs. The experiment data are derived from the representative ancient buildings in Forbidden City.
Experimental results show that the algorithm is able to manage more than 10G-level data and one billion valid points with
satisfactory rendering efficiency.
KEY WORDS: Terrestrial LIDAR, Point Image, Spatial Index, LOD
1. INTRODUCTION
1.1 Point Image
Terrestrial LIDAR has become one of the principal means of
acquiring fine 3D graphics information in the near earth space.
3D laser scanner launches the laser and ranges by receiving the
laser that had been launched for many times. The data are
acquired by high resolution domed scanning. The points are
recorded by 3D coordinates with intensity, which are stored by
the way of 2D array logically.
Figure 1. The domed scanning data: point image
We define original scanned data with the array way as point
image. The multi-view (multi-station) point images which are
only registrated are called point image model. Point image
model has the characteristics of large volumes (mass).
arrangement with scanning line(grid), raw data without any
processing(originality) , single point in the same rays
direction(uniqueness). Some terrestrial laser scanner equipped
with a camera can acquire texture information of scanning target.
There is no direction vector of points generally. However, the
rich information acquired by laser scanner has not been used
* Corresponding author. E-mail:guoming@bucea.edu.cn.
effectively. It is used to convert the data to scattered point cloud
and recalculate the geometric properties of scattered point cloud
again. The topological relationships acquired by scanning
directly among the 3D points are abandoned. Some scholars
have noticed this characteristic of point image model and have
made some meaningful work. In order to construct TIN model
more effectively, a algorithm of spherical directed search and
spherical Local Optimization Procedure(LOP) is presented,
which not only improves the efficiency of network
configuration, but also avoids triangulation error and hole
caused by the data noise(Zhang Fan, Huang Xianfeng,2009);In
order to optimize point image registration process, a spherical
re-sampling method based on distance dependence is used to
simplify point image data, which reduces the time cost of
registration effectively (Mandow, A., Martinez, J.L, 2010).
1.2 Research Significance
Point image can express true target shape as a whole. They are
original data that must be saved permanently, and the post-
processing work of point cloud data must depend on the point
image model. However, the volume of data is very high and
point image data may include millions of 3D points. It 13
obvious that point image model has lots of scanning stations.
For example, inside and outside the Hall of Supreme Harmony
in Forbidden City, the volume had been reached to hundreds of
stations. And how to store, manage and visualize such a huge
point image model has become a problem that must be solved at
present. Commercial software have functions of edit,
registration, 3D modelling and data conversion of point cloud
data, etc. But they are mainly used to solve the problems from
the traditional view of reverse engineering and can't render the
whole of point cloud data at once.These software can only
divide the data into sub-block to process, then choose the
available data and abandon the others, which loses the
originality and grid of point image data. It not only takes more
time cost and effort but also can’t be used conveniently.
Especially for the ancient buildings in Forbidden City, they
have feat
requireme:
organizing
In order tc
model of
and visuali
original c
mechanism
foundation
cloud data.
1.3 Prese
In recent |
cloud has
give a histc
Some rese:
organize px
2002a; Zhe
tree (Shi W
variants (Gi
From evol
representati
for various
etc, 2006)
multidimen
index (Robi
problem wl
overlap, Gt
presented tl
search pert
massive pc
proposed a
code and R
by Hilbert s
group under
improve the
and three-di
2009;Gong
spatial index
Detail(LOD
selection alg
Cauchy valu
distribution
Most spatial
of point ol
granularity c
object. It is
spatial data,
index tend t
external mei
backup typic
data. It uses
desired level
Systems ofter
index mecha;
is not a spat;
Shi Wenzhoi
Spatial index
2. SPATI/
21 The Des
The design o
of point imag
