In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds). 1APRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
2.2. Laboratory Work 
The main difficulty in the laboratory tasks resides on the data 
geometrical fitting. The large size of the object as well as its 
closed shape poses a major challenge on the use and 
optimization of the alignment and adjustment of the point 
clouds (Chen and Medioni, 1992). 
2.2.1. Alignment and adjustment of the laser scanner data. 
The alignment is done on a basis of an independent models 
orientation approach. Each consecutive pair of point clouds is 
oriented to each other in order to provide a set of relative 
orientation parameters. These parameters are then used as initial 
solution in a block adjustment procedure. More specifically, the 
alignment of each of the individual data sets is based on the 
solution of a rigid body transformation. As each set is expressed 
in its particular local system, at least three homologous points 
with the neighbouring sets must be provided. With this initial 
solution in which each cloud is referred to the first one, the 
original 3D transformation may be simplified, resulting the 
following equation: 
x r 
AX 
1 
-Ak 
Дф 
X' 
Y' 
= 
AY 
+ 
Ak 
1 
-Ato 
* 
Y 
Z' 
AZ 
-Дф 
Aco 
I 
Z 
where (X\Y’,Z’) are the coordinates in the new frame, {X, Y,Z) 
are the coordinates in the input frame, (AX,AY,AZ) and (co, <p, k) 
are the parameters of the relative orientation, translations and 
rotations, respectively. 
Afterwards, the alignment (2) will be refined by an iterative 
least squares adjustment in which all the points of the 
overlapping area will be involved. In addition, and with the 
idea of minimizing the closure error to half its value, two 
completely opposite clouds are used as original references in the 
initial alignment. The clouds are divided into two subsets 
instead of using a unique initial cloud as a reference for whole 
lot of them. 
Finally, once the independent model adjustment is completed, a 
block adjustment is launched. This procedure is completed by a 
network of GPS control points that has been designed and 
measured on the upper part of the walls. 
2.2.2. High resolution images registration. This phase consists 
on solving the outer orientation of each of the high resolution 
images. The parameters are referred to the laser scanner frame. 
It is a manual process in which the so called DLT (Direct Linear 
Transformation) (Abdelaziz and Karara, 1980) is used. This 
model is based on the popular collinearity equations (3). In it, a 
eleven parameter model relates the image coordinates (.v,v) to 
the object laser coordinates (X',Y',Z ) (see equation 2). In this 
way, all the images are referenced to the laser scanner data 
frame. 
(x r r„ -frn)X'+(x p r,, -fr l ,)Y'+(x | ,r J} -fruJZ'-Jx^rj, - fr n )X s '-(x p r, : -ft l; )Y s '-(x p r„—fr,j )Z S ' 
X_ r„X'+r,,Y'+ r„Z’-(r n X,'+ r,,Y s ’+ r v< Z,') 
(3) 
(y p r ?l - fi\,)X ’+ (y p r,, - fr 2 ,) Y ’+ (y p r„ - fr 2 ,) Z(y p r ?l - fi\, )X S '-(y p r 3i -fi^) Y s '-(y p r J3 -fiv, )Z s ' 
r 31 X'+ r 3; Y'+ r 33 Z - (r 3I X s '+ r 3; Y s '+ r 3 ,Z s ') 
The rest of the parameters of (3) are the singular elements of the 
rotation matrix (r,-,), the image coordinates of the principal point 
(x f) ,y p ) and the laser coordinates of the point of view 
(Xs’.Ys'.Z-s'). In (3) the distortion parameters are not expressed, 
as these parameters have been computed previously in a 
laboratory procedure and can be applied to correct the input 
image coordinates. Six homologous points, on each image and 
on the laser scanner point cloud, are identified and so, the 
exterior orientation of each image can be computed. More 
precisely, images captured from the blimp are registered taking 
singular elements (battlements) as homologous points in both 
dataset: point cloud and aerial images. This step is solved 
manually since the baseline and perspective between both 
dataset are radically different and thus the automatic registration 
could does not work at all. When these parameters are known 
the radiometric information of the image can be projected over 
the point cloud or over the triangle mesh by means again of the 
collinearity equations (3). 
2.2.3. Georeferencing. Finally, the whole data set is geo- 
referenced to the ETRS89 system. The chosen geodetic 
projection is UTM in the zone 30. In this way, it becomes 
feasible to integrate, in a simple way, cartographic and 
photogrammetric data. This last stage requires a field campaign 
to acquire the GPS observations but it is always possible to 
perform it at the same moment of the photographic or laser 
scanner campaign. Through the observation of three geodetic 
vertices and the geoid undulation model EGM08, the 7 
parameters of the Molodensky-Badekas (Welsch and Oswald. 
1984) can be computed. 
X' 
ДХ 
X c 
1 + dk 
8 Z 
-c Y ' 
pi-x c l 
Y' 
= 
AY 
+ 
Y c 
+ 
~ e z 
1+dX 
e x 
* 
Y - Y c 
Z’ 
AZ 
£ y 
-e x 1 + dk 
N 
1 
N 
where dX is the scale variation, (SxXyXz) are the elementary 
rotations and (X C ,Y C ,Z C ) are the coordinates of the centroid 
which is the origin of the rotations. 
3. RESULTS 
Table 1 gives a general vision of the large size of this project 
and the difficulties related to the acquisition and processing of 
the walls of Avila. We may stress that the wall heights range 
from 14.5 m at the east to 10.5m at the south; the towers height 
vary between 14 and 17 m. The towers of Puerta del Alcázar 
and Puerta de San Vicente exhibit a height of 20 m (Mariano 
Serna, 2002). 
Length 
2.516 m 
№ of towers 
87 
№ of Battlement elements 
(nowadays / original) 
2113 /2379 
№ of doors 
9 
Width of the wall 
Between 2.6 and 2.8 m 
Average height of the wall 
11.5 m 
Average height of the towers 
15 m 
Table 1. General information of the walls of Avila 
The ideal laser station is that in which the horizontal scan span 
covers two towers, the wall between them and part of the wall at 
the outer part of the towers so the alignment with the adjacent