> information as ID,
3B) and so on. Each
a point requires are
, à total of 24 bytes,
it needs one floating
oint object needs 32
ne of storing twenty
e it cannot be render
ry in reserved mode.
1 of the point image
data. Different data
n methods, for the
ve stretch their each
integers, using one
in the way of integer
retching method will
y; Inspired by 47th
antitative methods of
3D float type point.
is large, but it is in
iment, point quantity
al index structure is
ce between reading
index retrieval. With
ning point as | mm,
1 in space are at most
X, Y direction can
dinate using 14 bits;
nerally can't be more
express Z-direction
n 1000 m it will be
Similarly we use two
s shown in the Table
hod reaches 37.5%,
] nearly.
Compression
ital Bytes ratio
6 50%
1 25%
3 25%
2 50%
12 37.5%
f point image Data
ALYSIS
eractive visualization
trieval of the query
a personal computer
6600 2.4GHz. Main
GeForce8500GT. We
latabase management
el and database table
‚wer Designer. The
oped in ODP.NET
phic library. |
oint image model is
point image data ofa
M, including 358062
ring number of the
reen every time. We
realize this article's
sen introduced above,
ps; the right diagram
is a multi-block point image model data. The data file size is
269.14M. It contains 12 blocks of point images and 5347765
colorful points. The average frame rate is approximately 60 fps.
b
Figure 5 LOD display of point image (Left: Buddha's head)
and point image model (Right: Beihai Park)
The red wireframe represents the MBB property of node data.
The MBB size is related on the resolution. The higher the
resolution is, more intensive MBB wireframe is, and the higher
LOD level is, more details will be displayed in the computer
screen. What should be noted is that the calculation operation of
frame rate not starts when the graphics buffer is full but deal
with the rendering data forcedly and immediately. This method
calculating the frame rate is out of the limitations of the screen
refresh rate. True visualization speed can be reflected better.
What is reflected in Table 2 is that the efficiency of rendering
point image data. According to the size of point sets after the
LOD select the frame rate of valid point is different. Guarantee
the basic premise of rendering quality, the rendering speed meet
or exceed 10000000 (points/second) in our experiment. It is
more than 2 times faster than Reference 15(ZHANG
Long, ,ect,2005), and more than 6 times faster than Reference
16(L. Ren, ect, 2002). Total time of LOD data retrieval includes
multi-station frustum cutting, LOD selecting and occlusion
culling,
Row Total : Total
; Point ;
File number number time of
2 Data Number Frame
Size/ Blocks & of of actual d LOD rate/fps
M column effective ; Data P
: rawing :
number point retrieval/s
40000- 110-
18.52 530* 2 ;
1 530*789 35806 200000 0.04318 458
121*95
269.1 2963*60 200000-
3 5 ;
4 12 7 5347765 350000 0.78731 30-80
Sires
1682*86
421.6 $ 550000-
7 568*2 T -
0 se 76 7774564 1340000 0.74863 14-50
Table 2.Rendering efficiency statistics of point image model
data using the proposed algorithm
The algorithm that this article presents supports to render huge
Point image model data. If the size of point’s number is larger
than a threshold it will retrieval a certain level’s data from 3D
R-Tree and QMBB-Tree. So it is able to maintain a high frame
rate with acceptable rendering quality. As shown in the Figure 6
the left is whole point image model of the Hall of Supreme
Harmony in Forbidden City, the file size is 1.54G, it is
Composed of 241596602 effective points, the valid quantity of
3D points is 10711 195; the right is whole point image model of
Zhonghedian, the file size is 4.74G, it is composed of
2038308028 effective points, the valid quantity of 3D points is
3940690.
,
Figure 6 Overall visualization of point image models
4. CONCLUSIONS AND FUTURE DEVELOPMENTS
The algorithm can read huge point image data and construct
OMBB tree spatial index quickly at the same time. It accesses to
organized spatial index data and point cloud data with 3D R-
tree and QMBB-tree hybrid index. It can avoid unnecessary
LOD data backup in external memory and the redundancy of the
sampling points. It can save the space of external memory and
main memory. Its query efficiency is higher in visualization
operation and data structure is simple and effective. Under the
premise of basic rendering quality it achieves higher rendering
efficiency. The security of data and concurrency are ensured
while taking advantage of commercial database to store spatial
data. However, network bandwidth will limit transmission
speed of point image data. It maybe affects the speed of point
image model of real-time visualization.
With the rapid development of LIDAR data acquisition and
processing technology, the data amount of a variety of spatial
data and non-spatial data will continue to increase. So we need
to consider the distributed and grid database management.
Aiming to rendering quality and rendering efficiency a new
GPU programmable means and adaptive point image model
rendering mode should be used. At the same time, CUDA
parallel computing architecture will be developed and applied
for higher quality and efficient visualization systems.
References:
Zhang Fan, Huang Xian-feng, Li De-ren, 2009. Spherical
Projection Based Triangulation for One Station Terrestrial
Laser Scanning Point Cloud[J]. Acta Geodaetica et
Cartographica Sinic. Vol.38( 1).
Mandow, A., Martínez, J.L., Reina, A., and Morales, J,2010.
Fast range-independent spherical subsampling of 3D laser
scanner points and data reduction performance evaluation for
scene registration. In Proceedings of Pattern Recognition
Letters. 1239-1250.
ZHENG Kun, ZHU Liang-feng, WU Xin-cai, LIU Xiu-guo, LI
Jing. Study on Spatial Indexing Techniques f or 3D GIS[J].
Geography and Geo-Information Science. Vol.22 No.4, July
2006.
HE Zhen-wen, ZHENG Zu-fang, LIU Gang, WU Chong-long,
2011. Dynamic Generalized List Spatial Index Method.
Geography and Geo-Information Science. Vol.27( 5),pp9-15.
Li Qingquan, Yang Bisheng, Shi Wenzhong, Li Zhijun, Hu
Qingwu, 2003. The theory and methods of spatial data
acquisition, modeling, visualization[M], Wuhan University
publication. Wuhan: Wuhan University Press.
Shi Wenzhong, Wu Lixin, Li Qingquan, Wang Yanbing, Yang
Bisheng, 2007. Models and Algorithms for Three Dimensional
Spatial Information System[M ].Beijing: Publishing House of
Electronics Industry.
