
9. CONFIGURATION
9.3.2 Disparity Settings
If the operation mode is set to stereo matching, then the `disparity settings'
allow for a conguration of the disparity range that is searched by SceneScan.
The disparity range aects the frame rate that can be achieved. The frame
rate should be adjusted once the disparity range has been changed
(see Section 4.4 on page 5 for recommendations). Please be aware that in-
creasing the disparity range will also reduce the maximum image size that can
be congured.
The `number of disparities' option species the total number of pixels that
are searched for correspondences. This option has a high impact on the depth
resolution and the covered measurement range (see Section 8.2). The start of
the disparity range can be chosen through the `disparity oset' option. Typi-
cally, a value of 0 is desired for the oset, which allows for range measurements
up to innity. If the observable distance is certain to be constrained, then low
disparity values won't occur. In this case it is possible to increase the disparity
oset, such that these low disparities are not computed.
9.3.3 Algorithm Settings
The behavior of the image processing algorithms can be controlled through
the algorithm settings. The default conguration has been determined using
machine learning methods, and it should thus be the best choice for most use
cases. Nevertheless, all algorithm parameters can be adjusted through the web
interface. The following parameters control the stereo matching algorithm:
Penalty for disparity changes (
P
1
): A penalty that is applied to gradually
changing disparities. A large value causes gradual disparity changes to
occur less frequently, while a small value causes gradual changes to occur
more frequently. This value must be smaller than
P
2
.
Penalty for disparity discontinuities (
P
2
): A penalty that is applied to
abruptly changing disparities. A large value causes disparity discontinu-
ities to occur less frequently, while a small value causes discontinuities
to occur more frequently. This value must be greater than
P
1
.
SceneScan implements several methods for post-processing the computed
disparity map. Each post-processing method can be activated or deactivated
individually. The available methods are:
Mask border pixels: If enabled, this option marks all disparities that are
close to the border of the visible image area as invalid, as they have a
high uncertainty. This also includes all pixels for which no actual image
data is available, due to the warping applied by the image rectication
(see Section 8.1).
22