Diplomarbeit Till Sieberth
Topic:
Development of a method to differentiate between illuminated
and shadowed areas based on a 3D surface model
Description:
Currently, few
algorithms exist for shadow detection or correction with the
help of 3D surface models. For airborne applications, some
algorithms are available that work with digital surface models
(DSMs) to correct the intensity values in hyperspectral images.
However, with the increasing importance of terrestrial 3D
modeling, texturing and visualisation, the differentiation of
illuminated and unilluminated areas becomes more important. The
Centre for Integrated Petroleum Resesach
(Uni CIPR) works on geological outcrop analysis based on
terrestrial laser scanning integrated with terrestrial
hyperspectral line scanning. In this application the
identification, and eventual correction, of shadowed areas
is important for improving processing efficiency and data
quality in close range hyperspectral imagery.
Fig. 1: 3D outcrop model, coloured with true colour images (left) and hyperspectral classification result (main face), (Uni CIPR, 2012)
For an accurate classification of different rock types in
the outcrop, shadowed areas need to be excluded or corrected
before analysis is performed. The manual masking of these
areas, or detection based on classification or algorithms
thresholding is slow and prone to error. The aim of this thesis
is to develop a geometrically correct algorithm for separating
illuminated and unilluminated areas.
Fig. 2: Hyperspectral image reduced to a greyscale image, white lines equal calculated shadow borders
Additionally, the algorithm calculates the geometrically
correct transition between both areas, which is important
because of the high spatial resolution, much higher than in
airborne images. Supplementary the incidence and radiation
angle of every picture element (pixel) can be calculated and
combined into a mask for further correction of the intensity
values of each element.
Fig. 3: Overview of results (left to right: smoothed shadow mask, incidence angle, angle of radiation , intensity image)
The input values for processing are a 3D model, a time
value, image registration parameters and an image, making the
method also attractive for already existing data. Based on
these typical input data, an accuracy of better than one pixel
can be achieved for the generated mask.
Acknowledgements:
The Research Council of Norway and Statoil ASA are thanked for
financial support in data collection. Thanks to the Virtual Outcrop Geology Gruppe for use
of the data and project support.
