The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
In order to register accurately and efficiently, the process of 
registration is divided into two steps, primary registering and 
accurate registering. 
2.2 Primary Registering 
The first step is the primary registering with the geometry 
feature constraints. Signal points can be used as constraints in 
registering. We can also fit the spheres ^ planes and cylinders 
from the point cloud. Parameters in those geometries can be 
used as constraints in registering. 
In order to simplifying the calculating process, yant symmetric 
matrix M is used to construct rotation matrix. 
M = 
0 — c -b 
c 0 -a 
b a 0 
(2) 
T -X -ARX 
(7) 
2.3 Accurate Registering 
After primary registering, the data in different stations are 
transformed from independent coordinate systems into a unified 
coordinate system on the whole. The accurate registration is 
based on ICP algorithms and registering with features algorithm. 
In this process, directly searching method which uses the basic 
spatial cubes with optimal step size is put forward in order to 
make it efficient for searching the corresponding points in the 
3D space to gain the relative parameters accurately. At the same 
time, the coordinates of each original station are considered. 
The aim is to get a weight value for the parameters of each 
point, and to ensure the precision of registration and the high 
accuracy of the data after redundancy eliminating and merging. 
The process of accurate registration is shown in the Figure 1. 
Primary' Registering 
R = (I-M)-'(I + M) 
(3) 
Because matrix M can be easily proved to be orthogonal matrix, 
the rotation matrix R can be presented with a, b and c. 
R = 
If 
1 + a 2 -b 2 -c 2 
2c - 2 ab 
2b + 2 ac 
-2c- 2 ab 
\-a 2 +b 2 -c 2 
2 a - 2be 
-2 b + 2 ac 
-2a- 2 be 
\-a 2 -b 2 +c 2 
(4) 
X = (x,y,z) T ' 
X'=(x',y',z') T , 
T = (AX,AY,AZ) T 
The transformation equation is: 
X' =T + XTX (/U,) 
\çX = (<2, b, c) T after deduction, we can get: 
(5) 
Consider original stations 
Choose basic station 
4 
Divide spatial cubes and get 
corresponding points 
Get weight values 
Registration with parameters 
¿7/7: 
Divide spatial cubes and get 
corresponding points again 
I 
Precision evaluation 
N l 
Fulfil the error condition 
Y 
i 
Redundancy FJiminating 
X = (A T Ay x A T L 
where 
A = 
0 -z-Az' -y-Ay' 
-z- Az' 0 jc + Ax' 
y + Ay' x + Ax' 0 
(6) 
Figure 1 Flow Chart of Accurate 
Registration 
The research results of this part mainly include three aspects as 
follows: 
L = (x'-x y-y z'-z) T 
So the rotation matrix R can be obtained. At the same time, the 
translation vector T is given by: 
1. Selecting the optimal side length of the spatial cubes. 
The side length of the basic spatial cube is decided according to 
resolution of the points cloud. Different resolutions of points 
cloud are applied in the experiments. Firstly, get the maximal 
606