on the bottom of the Buddha and six of them were placed on 
overlapping regions. 
(3) Place the total station on the related ground control points, 
get all centre coordinates of artificial markers in the Beijing 54 
reference datum. All the artificial markers were scanned with 
high accuracy too, their point cloud data were collected and 
their centre coordinates in the standalone survey reference 
datum were gotten by special software. Then all the centres of 
the artificial markers will have a pair of coordinates, one is in 
the Beijing 54 reference datum and the other is in the local 
survey reference datum. 
(4) The transformation parameters were calculated on the basis 
of at least three corresponding pairs of coordinates with 
formula 1. If more than three pairs, least mean square will be 
employed to determine all the transformation parameters. Then 
all the point cloud data may be transformed into the Beijing 54 
reference datum by the calculated transformation parameters 
and formula 1. 
fxi 
f r 
A n 
r !2 
r 
M3 
(xi 
i%) 
Y 
= A 
r 21 
r 22 
r 23 
y 
+ 
Y„ 
y Z J 
v r 3i 
r 32 
f 33> 
N 
O 
where A = scaling parameters 
r n ~r 33 = rotation parameters 
x, y, z = object initial coordinates 
X 0 , Y 0 , Z 0 = shift parameters 
X, Y, Z = object transformed coordinates 
(5) Repeat the above work procedure and all point cloud data 
will be joined into the Beijing 54 reference datum one by one. 
4.1.2 Registration procedure with Method 2 
In this case artificial markers and extracted features will be 
used to calculated transformation parameters and register the 
continuous scans. Because the quantity of distinct features 
which can be extracted easily was small, several markers were 
placed on the bottom of the Buddha in order to improve the 
quantity of distinct features. Its work procedure lists as follow: 
(1) The local reference datum of scan 2 is selected as the 
reference datum and all point cloud data in other scans (scanl 
and scan 3) will be transformed into the local reference datum. 
(2) Extract at least three corresponding pairs of distinct features 
from the overlapping region of the two neighbouring scans and 
get their coordinates. 
(3) Least mean square will be employed to determine all the 
transformation parameters on the basis of the corresponding 
pairs of coordinates. 
(4) All the point cloud data in scan 1 may be transformed into 
the local reference datum of scan 2 by the calculated 
transformation parameters and formula 1. 
(5) Repeat the above work procedure and all point cloud data 
in scan 3 will be merged into the local reference datum of scan 
2. 
4.1.3 Registration procedure with Method 3 
In this case all registration work will be finished by special 
software after giving enough initial values, no artificial 
markers and extracted features are needed. Its work procedure 
lists as follow: 
(1) The local reference datum of scan 2 is selected as the 
reference datum and all point cloud data in other scans (scanl 
and scan 3) will be transformed into the local reference datum. 
(2) Three corresponding pairs of distinct feature points from 
the overlapping region of the two neighbouring scans are 
selected to calculated the transformation parameters and finish 
the coarse registration. 
(3) The iterative closest point (ICP) algorithm is used to finish 
the next fine registration in order to improve previous achieved 
results. 
(4) Repeat the above work procedure and all point cloud data 
will be joined into the local reference datum one by one. 
4.2 Analysis and Comparisons 
For the three different registration methods three standing 
registration tests were done and the registration result image is 
shown in Figure 1. 
Figure 2. The registered point cloud image 
In order to compare the different registration methods, a special 
program was developed to calculated the distance between two 
overlapping region. The calculating procedure is described as 
follow: 
(1) After getting the 3D transformation parameters a scan data 
can be merged into another scan, and here two registered scan 
data are stored into two separate layers. 
(2) In order keep the initial conditions (overlapping region size 
and the quantity of candidate points) same, an attribute tag is 
stored into every point attribute record. 
(3) Select one of them as reference layer and calculate the 
distance from any point on another layer to reference layer.