597 
REGISTRATION OF TERRESTRIAL LASER SCANNER POINT CLOUDS 
BY ONE IMAGE 
C. Altuntas*, F. Yildiz 
Selcuk University, Engineering and Architectural Faculty, Department of Geodesy and Photogrammetry, 
Konya/TURKEY - (caltuntas, fyildiz)@selcuk.edu.tr 
Commission V, WG V/3 
KEY WORDS: TLS, Point Cloud, Registration, Space Resection, Pose Determination 
ABSTRACT: 
Registration of point clouds in the same coordinate system is the most important step in processing of terrestrial laser scanner 
measurements. Used methods for registration of point clouds have required scanned overlap area betweeen point clouds or adequate 
points for every scan in common coordinate system. These procedures are required time and labor more than necessary. In this study, 
a method has been explained for registration of point clouds in the same coordinate system by one image, which covers the scan area 
of each point clouds. Essentially, the method is performed in two steps. The first step for the process is calculated the exterior 
orientation (rotation angles and translation components) of each point clouds with respect to image coordinate system, and image 
and each scan axes is maked parallel for the other. Translation vector is determined by difference of projection center coordinates in 
each scan coordinate system. Evaluate of accuracy of the method used here is done by beforhand selected control points coordinates. 
Seen in the procedure, the method can be used for registration of laser scanner data but accuracy of registration of the method has yet 
sufficiently. The accuracies of the registration can be increased with removed distortion and other errors. 
1. INTRODUCTION 
The terrestrial laser scanner (TLS) is a time of flight instrument 
which can directly acquire very accuracy and dense 3D point 
clouds in a very short time. These instruments are used 
frequently in cultural heritage documentation, 3D modelling 
and acquire 3D data. Instrument’s data is point cloud which has 
local cartesian coordinates. If too many scanning is required 
for 3D modelling of one object, every scanning must include 
overlaping surface with the other adjacent scan. 
Too many different methods have been used for registration 
point clouds. The most popular method which is used to 
registration of point clouds is iterative closest point (ICP). ICP 
algorithm developed by Besl and McKay (1992), Chen and 
Medioni (1992). Althought the method required intensive 
computation and time expense, the most used method. The ICP 
assumes that one point set is a subset of the other. Point pairs of 
nearest points search in two point sets. Rigit transformation 
parameters estimated by the point pairs, then transformation is 
applied to the points of one set. Procedure is iterated until 
Euclidean distance is minimum between point pairs. The values 
of estimate transformation parameters of each iterations has 
used for approximate value of the next iteration. ICP algorithm 
is put on estimation a 6-parameter rigid body transformation 
with scale factor 1. ICP mathematical model is required initial 
values of unknowns. If first values of transformation parameters 
is good estimated, new values are estimated approximate after 
5-10 iterations. 
The negative sides of ICP registration methods are highly time 
consuming search for the nearest point and too much iterarions. 
To tackle the exhaustive search problem and improve of ICP, 
different techniques of the method have been application (Sharp 
et.al.,2002; Johnson and Kang, 1999). Different applications of 
ICP is named related with used technique and data, for example 
Color ICP (Johnson and Kang, 1999), Geometric ICP (Liu, 
2006). There can be found too many ICP approaches in the 
literature (Rusinkiewicz and Levoy 2001; Chetverikov, 2002). 
Another mothod used to point cloud registration or 3D surface 
matching is least sequare 3D (LS3D) matching technique. The 
method is developed by Gruen and Akca 2005. LS3D method 
is applied as least suquare matching of overlaping 3D surfaces, 
which are sampled point by point using a laser scanner device 
or other devices. Proposed method matches one or more 3D 
search surfaces with a 3D template surface, minimizing the sum 
of sequares of the Euclidean distances between the surfaces. 
The geometric relationship between the conjugate surface 
patches is defined as a 7-parameter 3D similarity transformation. 
As scale parameter is 1, 6-parameters is estimated, which have 
been needed for transformation of conjugate overlaping point 
clouds. LS3D matching is required initial approximate value of 
transformation parameters. 
Also, registration of conjugate overlap point clouds can be done 
by 3D conformal transformation (Scaioni, 2002). A set of n 
models acquired by laser scanner have been transformed from 
intrinsic reference systems to a common reference system. It 
can be either select of reference coordinate system a one of the 
models or ground control point’s (GCP) reference system. For 
each situations, a minimum of three points is required to 
compute the registration between two adjacent scans. Initial 
values for the transformation parameters must be provided to 
compute of conformal transformation. Transformation 
parameters between adjacent scans will be computed by least 
sequare adjustment. Place and count of points on an overlap 
* Corresponding author.