TY  - GEN
UR  - https://archiv.ub.uni-heidelberg.de/volltextserver/25113/
AV  - public
A1  - Gutsche, Marcel
N2  - Computer vision plays an important role in the progress of automation and digitalization of our society. One of the key challenges is the creation of accurate 3D representations of our environment. The rich information in light fields can enable highly accurate depth estimates, but requires the development of new algorithms.

Especially specular reflections pose a challenge for many reconstruction algorithms. This is due to the violation of the brightness consistency assumption, which only holds for Lambertian surfaces. Most surfaces are to some extent specular and an appropriate handling is central to avoid erroneous depth maps. 

In this thesis we explore the potential of using specular highlights to determine the orientation of surfaces. To this end, we examine epipolar images in light field set ups. In light field data, reflectance properties can be characterized by intensity variations in the epipolar plane space. This space is analysed and compared to the expected reflectance, which is modelled using the render equation with different bidirectional reflection distribution functions.  

This approach allows us to infer highly accurate surface normals and depth estimates. Furthermore, it reveals material properties encoded in the reflectance by inspecting the intensity profile. 

Our results demonstrate the potential to increase the accuracy of the depth maps. Multiple cameras in a light field set up let us retrieve additional material properties encoded in the reflectance.
TI  - Light Fields
Reconstructing Geometry and Reflectance Properties
CY  - Heidelberg
ID  - heidok25113
Y1  - 2018///
ER  -