Graphics Reference
In-Depth Information
Fig. 5.16
Three-dimensional reconstruction results for a synthetically generated surface.
(
a
) Ground truth. (
b
) Noisy depth data. (
c
) Noisy intensity and polarisation angle images, based on
measured reflectance functions of a raw forged iron surface. The reconstruction result for noisy im-
ages of a surface with uniform albedo, obtained by SfPR with global optimisation without integra-
tion of sparse depth data, is shown in (
d
) using a single intensity image, in (
e
) using both intensity
images, and in (
f
) using one intensity and one polarisation angle image. (
g
) Reconstructed sur-
face obtained based on noisy sparse depth data alone. (
h
) Reconstruction result using sparse depth
data and intensity. (
i
) Reconstruction result using sparse depth data, intensity, and polarisation an-
gle. For comparison, the reconstruction result obtained based on SfPR with local optimisation and
without sparse depth data is shown in (
j
) using both intensity images and in (
k
) using one intensity
and one polarisation angle image
to be significantly lower for the intensity. The local optimisation approach accord-
ing to Sect.
5.3.1.2
provides a very accurate reconstruction for the noise-free case,
but performs worse than the global approach on noisy data (cf. Figs.
5.16
j-k). This
property can be observed clearly by comparing the corresponding reconstruction
errors of
p
and
q
given in Table
5.4
. With intensity and polarisation angle images,
Search WWH ::
Custom Search