592
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. P art B5. Beijing 2008
• Perform a first TLS acquisition that covers both the moving
and stable areas. In this stage it is important to consider that
the key parameters of the acquisition, like the sensor-to-
object distance and the sampling density, have a direct
impact on the deformation analysis results. Another key
factor is given by the distribution of the stable areas in the
observed scene. The best scenarios are given by the
deformation area completely surrounded by stable areas.
This aspect can strongly impact the quality of the global
matching, which in turn can generate systematic errors in
the derived deformation results. This can be relaxed by
using co-registration procedures based on a set of ground
control points (GCPs) or tie points distributed in the scene,
e.g. see Giussani and Scaioni (2004).
• After a given time period, which depends on the
characteristics of the deformation at hand, perform a second
TLS acquisition over the same area. Note that this operation
can be repeated several times, getting a complete temporal
deformation monitoring. We restrict the discussion to the
simplest case, based on two acquisitions. It is worth noting
that the geometry of the second acquisition can be slightly
different from the previous one. In fact, the procedure does
not assume to perform the data acquisition from the same
viewpoint.
2. Global Matching: We name hereafter master the first point
cloud, and slave the second one. Since each point cloud has its
own reference system, we need to co-register them. This is
achieved by applying a global LS3D matching over the areas of
the scene that remain stable between the two acquisitions. For
this purpose the parts of the scene that are subjected to
deformation have to be properly removed before performing the
co-registration. The deformation areas are often known from
external information on the observed scene. Otherwise some
extra processing is required to identify stable and deformation
areas. As already mentioned above, the global matching
represents a critical step of the procedure, which can directly
affect the estimation of the deformation parameters. For this
reason, the key characteristics of the studied area have to be a
priori properly considered, and the feasibility of the global
matching have to be assessed. In case of weak configurations
one should consider the option of performing a co-registration
based on GCPs or tie points. If the configuration is feasible and
the global matching is performed, its quality has to be checked
by taking advantages of the standard LS based tools, e.g.
analysis of the residuals, outlier rejection, etc.
3. Local matching: Once the two point clouds are co-registered,
the deformation analysis involves two key steps:
• Identification of the specific areas to be analysed. This can
be done by selecting on the master point cloud specific
objects or areas of interest, and can be properly performed
working on the TLS intensity data. Alternatively, if no
specific areas of interest have to be studied, the study area
can be divided in subsets that contain a specified number of
TLS samples. Note that this operation can be performed
through an automatic procedure.
• For each subset, using the LS3D matching, the
corresponding subset on the slave is automatically searched,
obtaining the transformation parameters (local matching).
These parameters, which in the experiments described in
this work are six, three translations and three rotations,
describe the deformation that each subset has undergone
between the two acquisitions.
The presented procedure has been validated using both a
simulated deformation scenario, which is described in the
following section, and a real landslide monitoring scenario; see
for in depth description Monserrat and Crosetto (2008).
3. DESCRIPTION OF THE VALIDATION
EXPERIMENT
In order to verify the effectiveness and evaluate the precision of
the proposed deformation measurement approach an experiment
was realized. In the experiment a deformation measurement
scenario was simulated. On this basis a comparison was done
between results estimated with use of two independent
techniques: the proposed TLS approach and a topographic
survey with total station. The experiment took place in the
Mediterranean Technology Park of Castelldefels (Barcelona),
where the Institute of Geomatics is placed.
Figure 2 shows the scene where the measurements took place
and the artificial targets in its bottom part. It is an intensity
image of the entire measured area. The scene includes some
buildings and structures in the back that were used as stable
areas. The numbered objects located on the bottom part of the
image are the 10 artificial targets that were displaced during the
experiment in order to simulate a deformation phenomenon.
The targets have an approximate dimension of 60 by 120 cm
and they differ one from the other for their geometry and the
type of material (wood, concrete, aluminium, foam, cartoon,
etc.). Figure 3 show the area selected as stable area for the
Global Matching.
Figure 2: TLS intensity image acquired during the experiment,
at an average distance of 134 m. Over the same image are
indicated the approximate directions of two of the three axes of
the reference system used for the data analysis. The Y direction
is approximately parallel to the TLS range direction
The measurements have been performed with use of the laser
scanner ILRIS 3D of Optech and Trimble 3601 DR total station.
The experiment was realized in the following main steps:
• Measurement of the scene with the TLS, acquiring the data
from two different positions:
o the first one, with an average distance sensor-object of
134 m, is called hereafter the 100 m dataset,
o the second one, with an average distance sensor-object
of 225 m, is called hereafter the 200 m dataset.