Full text: Mapping without the sun

into the control network well, then this one bad tie-in has no 
effect on the other scans being tied to the network. 
There are two ways to populate a survey network with scans. 
One involves using scanners that can be directly geo- 
referenced over points, similar to the way a total station is used. 
The other way is analogous to photogrammetry and needs to 
place known points or targets into scanner’s field-of-view. 
However the former way has low accuracy and the latter has 
higher accuracy. For any of them all the known controls and 
targets must have known coordinates in a uniform coordinate 
system, in some case it's a demerit and becomes unpractical. 
This registration methodology is most commonly used for long 
straight projects, such as roadways or site surveys. 
3.2 Method 2: Direct merging of scans with each Other 
This approach involves linking scans to each other dependently. 
In this case, scans must have some overlap. There are two 
commonly used methods for directly linking overlapping scans 
together: (1) Special scan targets are placed within the 
scanner’s field-of-view. Each target is scanned at high density 
and can get high-accuracy centre point coordinates by special 
software. Two scans that share identical sets of targets can be 
registered to each other. (2) It is not always practical to place 
targets (or enough targets) within a scanner’s field-of-view. In 
this case, certain specific features, such as a pointed comer or 
an edge of an object, may be visible from multiple scans. Such 
a comer or edge can be scanned at high density, just like a 
target. Using special software, comer scan data can be 
converted into a model and comer coordinates can be extracted. 
These comer points or vertices can be treated just as formal 
targets are treated to register adjacent scans to each other. This 
method is generally less accurate than target-based methods, 
but suffices for certain projects. 
3.3 Method 3: “Cloud-to-cloud” registration 
This is a more recent innovation to high-definition surveying, 
but it can be very handy and powerful if used properly. In this 
method, neither targets nor special features are processed into 
modeled vertices. Instead, clouds are aligned to each other by 
selecting three or four "common points” within the overlap 
area of each cloud. These common points are selected to be 
physically close to representing the same point within each 
overlapping scan. Special software is then used to align the 
entire surfaces of the overlapping scan clouds to each other. 
This method is appealing because it reduces the need for 
placing targets in the scanner’s field-of-view and 
scanning/surveying them. 
In the right conditions, “cloud-to-cloud” registration can 
provide amazingly accurate overall results. Rather than relying 
on a dozen or so target-based points for network adjustment, 
cloud-to-cloud registration may literally be taking advantage of 
a best fit based on hundreds of thousands of points. 
Till now, much work has been done on the “cloud-to-cloud” 
registration. One of the most popular methods is the iterative 
closest point (ICP) algorithm developed by Besl and McKay 
(1992). Several variations and improvements on the ICP 
method have been shown (Masuda and Yokoya, 1995; 
Bergevin et al., 1996). The iterative closest compatible point 
(ICCP) algorithm has been proposed in order to reduce the 
search space of the ICP algorithm (Godin and Boulanger, 1995; 
Godin et al., 2001). 
Because Yungang Grottoes has the huge quantity and scatters 
on a wide region, different Grottoes have different work 
environment, and their data collecting procedure and data 
processing procedure will be different to some extent. In order 
to get the common registration method representative Grottoes 
were selected to make the test. And Stone Buddha No. 20 was 
selected after fully analyzing all kinds of conditions. 
4.1 Point cloud data collecting and registration 
Stone Buddha No. 20 is tall and wide so its whole point data 
can not be gotten on a standalone position by TLS which has 
limited view field. In order to get the whole data of the Buddha, 
it’s necessary to work at different positions and from different 
views. In data collecting procedure, in order to construct whole 
3D model of the Buddha successive scans must be keep more 
than 15 percent overlapping region. In this test Optech ILRIS- 
3D which is made in Canada was employed to scan the Buddha, 
and we made nine scans from three standalone positions (from 
the left, the front and the right, named as scan 1, scan 2, scan 3) 
and nine different views. In order to keep the registration 
quality more than 50 percent overlapping proportion were kept 
between the contiguous scans. Three scan bounds and their 
overlapping bounds related to three different standalone survey 
positions are displayed in Fig. 1. Moreover, images were gotten 
at the same time in order to get the real textures and construct 
the real 3D scenes. 
Figure 1. Three scan bounds and their overlapping bounds 
4.1.1 Registration procedure with Method 1 
In this case every scan is joined into a uniform reference datum 
independently, so no overlapping regions are need between 
neighbouring scans. Its work procedure lists as follow: 
(1) Based on Beijing 54 reference datum a common control 
network was established for Yungang Grottoes, and related 
control points were placed near Stone Buddha No. 20. 
(2) All point cloud data were gotten from three different 
standalone survey positions, and more than 50 percent 
overlapping proportion were kept between the neighbouring 
survey positions. Because the Buddha is too high to place 
artificial markers on the top, 10 artificial markers were placed

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.