Here you will find selected experimental results of my computer vision research.
Please, feel free to
contact me for experience exchange.
Research objective
Since the middle eighties, there are effective algorithms to obtain a three-dimensional
model from digital images of a rigid scene. Today, nobody is surprised by reconstructed
dense surfaces with a high detail level even for non-rigid scenes. However,
for many practical optical measurement tasks, the contemporary established methods
do not provide sufficient accuracy and precision.
My work on computer vision focuses on getting as much properties of a
three-dimensional scene as possible from only as small number of images
as necessary. In particular, I work on the following problems:
-
Recovery of (sufficiently) smooth image intensity functions from discrete digital image data
-
Joint models for image interpretation, reconstruction, and camera calibration
-
Quality assessment of the reconstruction results
Reference data for stereoptic, trioptic, or multiple view reconstruction, in particular of humans,
is rare on the internet. I decided to contribute with my own reference sets which I use to evaluate
my research results.
Setups
The data sets are created with one of the two setups: the linear table and the camera box.
The linear table is for smaller objects as well as for accurate adjustment of the
camera position. The linear table admits accurate displacements at a resolution of 0.1mm.
The box setup is designed to acquire synchronous multiple view image sets of human heads. Typically,
i use it to obtain trioptic synchronous picture sets.
Reconstruction data
The files below are a starting point and there are more to come as I proceed
with publishing my work.
The input set contains 3 asynchronously taken images of a PCI board with a linear table.
The motion between the images is a pure translation with a calibration available.
The camera is equipped with a polarization filter to reduce specularities.
The output set contains a facet reconstructed from the three images.
The input set contains 3 synchronously taken images of the person.
The output set contains a facet reconstructed from the three images.
The input set contains 3 synchronously taken images of the person.
The output set contains a facet reconstructed from the three images.
Here you will find the
format specifications
for custom resources, like z-images.