The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
593
• Measurement of the coordinates of 15-20 marked points per
each target. The points were measured manually with a
precise reflective prism and Trimble 3601 DR total station.
In addition, other targets located on the buildings and
visible in Figure 1 were measured in order to guarantee a
set of tie points that link the TLS with the topographic data.
• Movement of the 10 targets, imposing displacements with
different magnitudes (approximately between 19 and 60 cm)
and directions.
• Repetition of the steps 1 and 2.
• Estimation of the 6 parameters of deformation for each
target by using the proposed TLS approach: Global
matching over the stable areas (the scene shown in Figure 1,
with the exception of the 10 targets) and Local matching
over each target.
• Independent estimation of the 6 deformation parameters of
for each target by using the topographic data.
• Comparison per each target of the two independently
estimated sets of 6 parameters of deformation.
• Analysis of the results. The outcomes of the analysis are
discussed below.
Figure 3: Data used in the global matching of the point clouds.
3.1 Validation results
Once the experiment was done the comparison between the
results coming from the proposed TLS approach and the results
coming from topographic survey was performed. The
topographic results are treated here as the real values here since
their precision is assumed to be one magnitude better than the
precision of the results coming from TLS. The differences
between the TLS estimations and the estimations coming from
topography represent the TLS errors.
Results are compared for both 100m and 200m datasets. For this
analysis it have been used only the translations. More detailed
analysis can be seen in Monserrat and Crosetto (2008). The TLS
errors for the three estimated movement components (X, Y and
Z) are shown in Table 1. Note that due to some occlusions that
had place in the area in the dataset acquired from 200m only 8
targets were measured.
The validation results of the 100 m dataset are summarized
below:
100m
T1
T2
T3
T4
T5
T6
T7
T8
T9
T10
X
0.6
0.2
0.8
0.6
0.3
1.1
2.0
0.5
0.1
-0.6
Y
0.5
0.5
0.1
0.8
1.0
0.9
1.1
1.6
0.2
-1.3
Z
0.3
0.5
0.8
0.4
1.2
0.3
0.6
2.4
1.9
-0.3
200m
X
0.8
-
-
2.4
0.3
5.7
2.7
2.2
1.0
1.1
Y
2.3
-
-
1.3
2.6
1.0
1.0
1.4
1.8
-1.8
Z
0.4
-
-
0.5
1.3
4.4
1.4
2.8
2.9
-2.4
Table 1: TLS errors in the estimated deformation displacements
(X, Y and Z) for the 100 m and 200 m datasets. The errors are
given in centimetres.
• in X, 8 of 10 targets have errors with magnitude below 1 cm;
the mean absolute error over the 10 targets is 0.7 cm, and
the maximum absolute error is 2 cm,
• in Y, 6 of 10 targets have errors with magnitude below 1 cm;
the mean absolute error is 0.8 cm, and the maximum
absolute error is 1.6 cm,
• in Z, 7 of 10 targets have errors with magnitude below 1 cm;
the mean absolute error is 0.9 cm, and the maximum
absolute error is 2.4 cm.
The 200 m dataset has the following characteristics:
• in X, 2 of 8 targets have errors with magnitude below 1 cm;
the mean absolute error over the 8 panels is 2 cm, and the
maximum absolute error is 5.7 cm,
• in Y none of the 8 targets has an error with magnitude
below 1 cm; the mean absolute error is 1.7 cm, and the
maximum absolute error is 2.6 cm,
• in Z, 2 of 8 targets have errors with magnitude below 1 cm;
the mean absolute error is 2 cm, and the maximum absolute
error is 4.4 cm.
Results presented above indicate significantly better
performances of 100 m dataset in respect to the 200 m one.
Secondly it can be also observed that there are no remarkable
differences between the errors associated with the three
displacement vector in X, Y and Z. In the following section a
detailed analysis of the above data is done.
4. CURVE MATCHING
The first results of the deformation measurement procedure
described above were achieved only using the surface matching.
The procedure has been recently extended to include the curve
matching procedure proposed in Gruen and Akca (2005). The
proposed approach involves the extraction of contours from
objects in the given point clouds and the matching of these
contours. Below the main steps are described:
a. Identification of the same objects in the different point clouds.
b. Extraction of the contour of each object and for each point
cloud. In this way we get the curves to be matched. This step,
which is currently performed manually, needs to be
improved. For this purpose different kinds of edge detectors
could be used. However, there are limitations to get a fully