AN ACCURATE REGISTRATION METHOD BASED ON POINT CLOUDS AND 
REDUNDANCY ELIMINATION OF LIDAR DATA 
Yumin Li *, Yanmin Wang 
School of Geomatics and Urban Information, Beijing University of Civil Engineering and Architecture, 1# 
Zhanlanguan Road, Beijing, 100044, China 
Wgs-PS WGV/3 
KEY WORDS: LIDAR, Point cloud, Registration, Simplification, Adjustment, Accuracy analysis 
ABSTRACT: 
A new method for accurate registration and redundancy elimination on point clouds is presented in this paper. Firstly, the accurate 
registration arithmetic which based on Iterative Closest Point (ICP) algorithm and feature registration method provides an accurate 
way in the process of registering. And the redundant data in the point cloud after registering can also be deleted efficiently with an 
algorithm considering the weight value of each corresponding point. Furthermore, a 3D application system is developed in VC++6.0 
and OpenGL 3D graphic library. Thus, those problems in registration and redundancy elimination of point clouds can be solved 
efficiently in this system. 
1. INTRODUCTION 
With the development of the modem surveying technology, the 
3D laser scanning technology has been a hot destination of 
research in this field. MilLions of spatial points can be obtained 
within a second as the result of the improvement of accuracy 
and velocity (Yanmin Wang, 2007). However, large number of 
points becomes redundancy after the registration among some 
stations. It has caused a lot of problems in the process of 
storage % translating > displaying and so on. More over, the 
precision of the data analyzing and the quality of the surface 
reconstruction can also be impacted by the mass redundancy 
data. However, the existing algorithms in some correlative 
software have many problems in the data processing. For 
example, when the number of constraints is limited, or some of 
constraints were moved a short distance without our notice or 
something else happened in the process of scanning, the 
veracity of registration can not be ensured. At the same time, 
many points are redundant after registration. But whether this 
part of points should be deleted or which part of points are more 
accurate, there will cause a lot of error because of factitious 
factors. So a further accurate registration followed by 
redundancy elimination after the primary registration should be 
actualized as soon as possible. 
2. RESEARCH ON REGISTRATION 
2.1 Research on existing algorithms of Registration 
Registration plays an important part in 3D model acquisition 
and object recognition. If we want to obtain a whole model of 
an object, this object should be scanned by the laser scanner 
from many view point. The data from each scan station has its 
own coordinate system. 
X 1 
i* 2 l 
p 
p 
•yp 
> = R< 
kJ 
0) 
In equation (1), x 1 , y 1 , z 1 are the values of 3D coordinate of one 
point in the first scan station, x 2 , y 2 , z 2 are the values of 3D 
coordinate of one point in the second scan station. R is the 
rotation matrix and T is the translation vector. 
After registration, those data in different coordinate systems can 
be transformed into a common coordinate system. So the most 
important problem is get the rotation matrix R and translation 
vector T accurately. 
There are two main methods on data registration nowadays. A 
popular method for aligning two point clouds is the Iterative 
Closest Point (ICP) algorithm (BESL.P.J, 1992). This algorithm 
starts with two point clouds and an estimate of the aligning rigid 
body transform. It then iteratively refines the transform by 
alternating the steps of choosing corresponding points across 
the point clouds, and finding the best rotation and translation 
parameters that minimizes an error metric based on the distance 
between the corresponding points. The methods for searching 
closest points contain point to point searching method> point to 
plane searching method> point to projection searching method 
and so on. 
The other method is registering with features. This algorithm is 
based on features in the original data or the features which are 
reconstructed into geometries from the original points. These 
features can be points> spheres^ planes and cylinders. 
Corresponding author. Yumin Li: hello_lym@yahoo.com.cn.