The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
automatic procedure because some areas should be removed
manually, e.g. the vegetated areas where the TLS point
clouds, from the viewpoint of deformation measurement, are
particularly noisy.
c. Matching of the extracted curves. We apply the least squares
curve matching in order to estimate the transformation (i.e.
deformation) parameters.
The above presented approach can be used in two different
ways. The first one is to check and refine the global matching
based on surface matching. The second one is to provide good
initial parameters to the surface matching. In the other side, it is
worth to underline one of the critical points of the procedure
which is the extraction of the contours in an automatic way.
Here below are presented the preliminary test results obtained
over simulated data in order to check the capability of the
method and to improve it. Future works will involve the
validation of the proposed approach using real data. The first
test was done using a simulated curve, which is shown in Figure
4. This curve provides good geometric information in all
directions, i.e. it provides the geometric information needed for
solving all the transformation parameters. For the test two
different point clouds, of about 600 points, of the same curve
have been simulated. Then a Gaussian noise with standard
deviation of 1 cm has been added to each point cloud. The tests
consisted in applying a known 6-parameter transformation to
one of the two curves, and then to estimate the transformation
parameters by using the curve matching.
Table 2 shows the results obtained in three different tests. Each
table represents the mean and the standard deviation of the
differences between simulated and estimated parameters, which
were obtained using 30 different simulations. For the first table,
Tl, the rotation angles have been fixed to zero, i.e. the
estimated rotations are expected to be zero. The translations in
X, Y and Z direction ranged randomly between -10 and 10 cm.
The results of this test show that for the translations there is a
mean difference up to 3 mm, with a standard deviation of the
differences of about 1 mm. For the rotations the standard
deviation of the differences is up to 0.2 gons.
Z(m)
Figure 4: Simulated curve for testing and improving the
implemented curve matching.
For the second test, whose results are shown in table T2, the
simulated translations in X, Y and Z ranged between -10 and 10
cm, and the rotations angles between -10 and 10 gons in each
direction. The results show again small biases for the
translations. In X and Z the differences have the same
magnitude than in the previous. In the Y direction the mean
difference equals 7 mm. The standard deviation of the
differences has a noticeable increase, up to 8 mm. A similar
behaviour can be observed for the rotations.
Finally the last test, whose results are shown in table T3, has the
worst results. In this case the simulated translations in X, Y and
Z ranged between -20 and 20 cm, and the rotations angles
between -20 and 20 gons in each direction. One may notice that
the standard deviations of the differences are up to few
centimetres for the translations, and up to 2.3 gons for rotations.
These larger values, compared with the previous tables, are
probably due to the relatively large translation and rotation to
be estimated. In fact, in all simulations the approximate values
for the matching were set to zero.
Tl
Trans, of 10 cm
Simulated - Estimated Param.
(cm)
(gons)
Tx Ty Tz
Rotx
Roty Rotz
Mean
0.2 0.3 0.3
-0.1
0.0 -0.2
Stdv
0.1 0.1 0.2
0.2
0.1 0.0
T2
Trans, of 10 cm and Rot. 10 gons
Simulated - Estimated Param.
(cm)
(gons)
Tx Ty Tz
Rotx
Roty Rotz
Mean
0.2 0.7 0.3
-0.3
0.5 0.4
Stdv
0.8 0.6 0.8
0.4
0.8 0.7
T3
Trans, of 20 cm and Rot. 20 gons
Simulated - Estimated Param.
(cm)
(gons)
Tx Ty Tz
Rotx Roty Rotz
Mean
0.9 0.1 0.2
3.2 4.0 3.0
-0.8 1.2 0.6
1.2 2.3 1.7
Stdv
Table 2: Main statistics of the three tests done in order to
analyse the capabilities of the curve matching. Tl is related to a
test where no rotations were simulated. In T2 we added
rotations between -10 and 10 gons. In T3 the translations vary
between -10 cm and 10 cm, and the rotations between -20 and
20 gons.
5. CONCLUSIONS
A procedure for deformation measurement using TLS data has
been presented. The procedure is based on point cloud surface
matching algorithm described in Gruen and Akca, (2005). In
addition, the first results obtained with the proposed procedure
have been presented and a research topic concerning to the
preliminary results obtained with curve matching have been
discussed.
From the results of the validation experiment the following
aspects have been highlighted:
• In the 100 m dataset the majority of the targets have errors
below 1 cm in the three components of the deformation
vectors. Considering the non optimal characteristics of the
targets, these represent promising results.