User manual
Version: v4.5.2
Registration mean error
The registration mean error sets a tolerance on the error returned by the
marker
registration algorithm. The computed error is a measurement of the deformation between the re-
constructed
marker
and the theoretical geometry, linked to the accuracy of the positioning of the
marker
.
The bigger the registration error, the looser the requirement on the system accuracy.
Matching maximum missing points
The matching maximum missing points simply sets how many
fiducials
can be missed for a
marker
to be reconstructed by the system. When set to zero, every
marker
candidate for which a
fiducial
cannot be associated (due to the matching tolerance for instance) or is missing
(e.g. occultation) won’t be available in the frame data.
Tracking range
The tracking range sets how far the
marker
tracking engine is allowed to look for a
raw
data
when trying to find a
marker
from its last known position. This distance is measured in pixels, and must
be adapted to the typical movement speed of the
marker
in real usage and the application frame rate.
Tracking time span
The tracking time span sets over how many frames the tracking engine is allowed
to look for a
raw data
when trying to find a
marker
from its last known position. As the
marker
are only
reconstructed when a call to
ftkGetLastFrame
is performed, frames might be skipped if the application
refresh rate is lower than the device acquisition frequency. The tracking time span can be seen as a tolerance
on the number of skipped frame.
Edge blob detection
The edge blob detection option is a binary option, enabled by default. When it
is on, the
raw data
located at the edges of the sensor are rejected, because their computed position (i.e.
the centroid) are known to be biased by the ‘missing part’ of the raw data. This has then an impact on the
reconstructed 3D position of the fiducial, leading to a wrong estimation of the position and orientation of the
marker
.
Pixel weight for centroids
The pixel weight for centroid is another binary option, and is also enabled
by default. When it is on, the centroid of the
raw data
is determined taking into account the intensity of the
pixel to achieve a better estimation of the centroid position.
Advanced centroid detection
The advanced centroid detection is a binary option as well. When it is
turned on (the default setting), the advanced centroid detection algorithm is used. This algorithm allows to
decrease the jitter on a reconstructed 3D position by 20 % to 50 % with respect to the commonly described
method. Advanced centroid detection only works with pixel weight for centroid activated.
7.9 Phantom fiducials
A phantom fiducial is an artefact created by the triangulation algorithm, caused by pairing a wrong combi-
nation of
raw data
. This can happen when several
fiducials
are located on the same epipolar line. At the
time of the triangulation, there is no way to know if a reconstructed fiducial is a real one (i.e. corresponding
to a physical fiducial) or a phantom.
fiducials
sharing a raw data have a probability smaller than 1 (one): if a
raw data
is used
n
times, the probability of the resulting
fiducials
is set to
1
n
. During the registration stage,
if any of the ‘clones’ is used to reconstruct a
marker
, its siblings are removed from the data.
Atracsys / 2020-06-17 / 16:48:00
43 / 113
