International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
sensor and in concentric circles of constant view azimuth angle
for frame sensors. In addition, those circles have to be split up
in segments for different relative azimuth angle.
Due to missing sampling redundancy physical models cannot be
used and are impractical for the large variety of surface types in
a typical aerial image. So it is favourable to use linear semi-
empirical kernel models like those developed by (Wanner et al.,
1995). The inversion of linear models is a least squares
regression and results in a simple matrix inversion while models
with non-linear parameters would require calculation-intensive
adaptive inversion algorithms. For the correction step which
requires many forward calculations, simple kernel functions are
preferable.
As suggested in (Beisl et al., 2004) even a simple 3-parameter
Walthall model (Walthall et al., 1985) without distinguishing
between different ground types (“global correction”) shows
good results (eqn 11). There is an extended version including a
varying sun zenith angle (Nilson and Kuusk, 1989) (eqn 12).
p(8,, 9) 2 a6? - b0, cos pc (11)
0(8,0,,9) 2 a0?8? - b(8? 9?) - c0, cos p d (12)
where p = reflectance
6; = incident illumination zenith angle
0, = reflection view zenith angle
9 — relative azimuth angle
a, b, c, d = free parameters
Since the Walthall model does not include a hot spot term a
simple empirical elliptical kernel function (eqn 13) is added to
eqn (11) and (12) which is inspired by the hot spot distance
function of the Li-kernels from the AMBRALS model (Wanner
et al., 1995).
D = Jtan? 6, + tan’ 6, - 2 tan 6, tan O, cos @ (13)
In case of frame sensors and for reasonably short line scan
images the incident illumination zenith angle is constant for a
single frame, so there is no need to consider this angle in the
BRDF correction and eqn (11) can be used.
However, to cover larger areas, images are acquired in blocks
with large overlap (60 % - 80 %) for stereo measurements. In
order to make use of the redundancy and to ensure the proper
radiometric matching of consecutive images a sliding window
technique can be used by sampling the current image together
with the previous and the following image and invert this set of
samples to give the modelling function for the middle image.
Depending on the block size, neighbouring flight lines may
have considerable time offsets due to the flight planning
schedule and therefore require considering the sun zenith angle
as modelling variable.
(Chandelier et al., 2009) and (Hernandez-Lopez et al., 2011)
suggest sampling on a regular grid of so-called radiometric tie
points, followed by an adjustment process and call the
procedure "radiometric aerotriangulation".
A first implementation will contain an NDVI-based land mask
algorithm that prevents water areas from being sampled. This is
because the water BRDF is of a specular reflectance type which
is contrary to the land BRDF which is of a hot-spot type.
2.3. BRDF correction: Since the reflection process is a
linear function of irradiance, a multiplicative correction by the
ratio of the model values at the final geometry to the model
values at the original geometry is used.
P.(0,.9)= p(0,)*(R(6.,0)/R(0,,)) ^ a
where p, p. = observed and corrected reflectance
R(0, o) = modelled reflectance
0,= view zenith angle
6. = correction view zenith angle
@ = relative azimuth angle
3. CONCLUSIONS AND OUTLOOK
This paper has given an overview of practical methods to
correct for radiometric distortions in photogrammetric images
caused by environmental effects. The idea is to include as much
physical information as is available into those corrections in
order to give a true copy of the reality as if it were seen from the
ground. This information includes absolute radiometric sensor
calibration, solar position, and haze information.
As a future step, measured ground spectra can be used to
perform an in-flight calibration to improve the absolute
radiometric calibration for remote sensing purposes (i.e adjust
the calibration factors such that the measured spectra match
with the spectra of the corresponding atmospherically corrected
and reflectance calibrated pixels).
Furthermore a class specific BRDF correction should be
implemented to better adapt to the specific surface properties.
Therefore a proper classification has to be made with a special
treatment of the class boundaries.
The atmospheric correction could be improved with the
correction of the local adjacency effect to enhance the contrast
in the image and also include to correction of topographic
effects by varying terrain height, surface tilt and change in
diffuse illumination by the percentage of visible sky. Also a
shadow correction would be a favourable, but challenging add-
on.
However, the guideline for the implementation of any new
feature must be the operational and efficient processing, and
that no new artefacts are introduced.
4. REFERENCES
Beisl, U., and Woodhouse, N., 2004. Correction of atmospheric
and bidirectional effects in multispectral ADS40 images for
mapping purposes. In: Int. Arch. Photogramm. Remote Sens.,
Istanbul, Turkey, Vol. XXXV, Part B7, 5 pp.
Beisl, U., 2006. Absolute spectroradiometric calibration of the
ADS40 Sensor. In: Int. Arch. Photogramm. Remote Sens.,
Paris, France, Vol. XXXVI, part 1, 5 pp.
Beisl, U., Telaar, J, and Schônermark, M.v., 2008.
Atmospheric correction, reflectance calibration and BRDF
correction for ADS40 image data. In: Int. Arch. Photogramm.
Remote Sens., Beijing, China, Vol. XXXVII, part B7, pp. 7-12.
Beisl, U., and Adiguezel, M., 2010. Validation of the
reflectance calibration of the ADS40 airborne sensor using