ul 2004
red and
ere the
than c,
ing the
ient. the
1e total.
80% of
d black
224mm
intre),
right).
ve any
etation,
he left)
| single
n these
n has 18
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
own reference system. The reconstruction of the 3D model of
the surveyed object than requires the registration of the scans
in a unique (local or general) reference system.
This phase can be performed in an interactive environment
through the identification of the homologous points (e.g.
corners) inside the overlapping portion of two adjacent scans.
Once the points (at least three) are collected, a simple 6
parameter transformation (3 rotations and 3 translations) can
be estimated and all the points of a scan can be changed into
the coordinate system of a scan that has been assumed as the
reference system.
If this simple approach is followed, all the scans that describe
an object can be referred to a single coordinate system (e.g.
the first scan coordinate system or an external system defined
by at least three points).
The collection of tie points and the search of the homologous
inside the adjacent scans can be performed automatically
using reflecting targets. The reflecting target must be placed
on the object in such a way that at least three targets can be
found in the overlapping portion of two adjacent scans. The
size of the target must be large enough to allow the laser
scanner to record it and their displacement has to avoid the
alignment of them.
Figure 7. Example of data before alignment (above) and after
alignment (below).
The LSR 2004 automatic register unit has been developed
using the following criteria.
Let us consider two adjacent overlapping scans. Some
reflecting targets (at least three) have been placed in the
overlapping portion and recorded by the laser scanner. All
these targets can easily be automatically found (see par. 3.2)
and their coordinates can be recorded.
The purpose of this procedure is to connect each point of the
right scan to the homologous point of the left scan; the
reference system of the left scan (for example) is fixed and
only the points recorded by the right scan can rotate and
translate in space. :
In each scan 2 markers that have the same distance in
between (only avoiding the instrument tolerance) are chosen.
These first 2 markers (in each marker point set) are used to
define a spherical coordinate system. Using this reference
system the spherical coordinates of the remaining points are
then determined. A comparison between these two sets of
spherical point coordinates is carried out evaluating the
relative geometric distribution. The result of the comparison
is a number of connection points that can be found in the two
identified sets of markers.
This procedure is iterated for all the others possible 2 points
combination. The solution that generates the greatest number
of connection points is considered the searched one.
If the number of the identified homologous points is greater
than 3 (the minimum conditions for the registration of two
models) and if they are not aligned, it is possible to proceed
with the registration, otherwise the procedure ask for the
operator intervention.
The estimation of the six parameters of a 3D coordinate
transformation is performed using the accepted homologous
points. All the points of the right scan can then be transferred
into the reference system of the left scan using a well known
rototranslation model.
At the end of the process, the rototranslation parameters of
the point cloud (3 rotations and 3 translations in the space),
the identified connection point coordinates, the rototranslated
point rejects and the route mean square of the estimation are
shown in a specific window in order to give to the user the
possibility to test the quality of the automatic procedure.
Using LSR 2004, it is also possible to georefer a 3D model in
an external reference system. To do this, the user has just to
know the markers coordinates that can be found in a scan,
even in an external reference system. The homologous points
extraction and the point cloud 3D rototranslation occurs in
the same way as described for the alignment of two adjacent
scans.
From a practical and economic point of view is interesting to
determine the minimum overlap between two adjacent scans.
It is easy to understand that the precision of the registration is
directly correlated to the “distance” between the alignment of
the tie-points and their true location. As a consequence, the
minimum area circumscribing the optimal tie-point location
can be interpreted as the minimum required overlap between
two adjacent scans.
In order to define the minimum overlapping area between
two adjacent scans, some tests have been performed.
We considered in particular couples of adjacent scans with
overlaps ranging from 90% to 10%. In each case three tie-
points and three check-points (not used for the estimation of
the transformation parameters) have been signalised.
Analysing the discrepancies of the check-points it is possible
to demonstrate that a minimum of 30% of overlap will assure
a final precision comparable to the instrument accuracy.
In this case is not necessary to provide control points and the
registration can be performed in the reference system of one
of the two involved scans.
The necessity to refer the survey to an external reference
system can be performed at the end of the process using the
procedure described in par. 3.4.
3.5 Laser scanner triangulation
Very large object and/or complex shape of the object require
the multiscan registration. If one tries to register all the scans
using the above described algorithms at the end of the
process unexpected and unacceptable deformation of the final
3D model arises.
Simply connecting 8 scans with a 30% of overlapping on a 40
m length fagade the discrepancies tested on a set of check
