Available Student and Research Projects
Contents |
Student Projects
We always offer projects to motivated students (first degree, Masters of Science, ERASMUS student, ...). If you are interested in doing a project in computer vision, let us know. Here is a description of ongoing research areas to give you an idea of possible projects. You can also suggest a research topic yourself.
If you work with us, you can learn a lot of skills which are relevant for a career as a software developer:
- Computer Vision, Signal processing, Robotics
- Linear Algebra, Analysis
- Software Engineering
We are using state-of-the-art platform-independent software tools:
- Source-code documentation with
doxygen
- Cross-platform user-interfaces with
Qt. You can develop full-featured GUI-software which runs under 
GNU/Linux, 
Microsoft Windows, and 
MacOS!
- Platform-independent
Standard Template Library
- Platform-independent
Boost Library
Project areas
Stitching for microscopes
 |
Premise
- A microscope-video of an object being moved in x-, and y-direction (parallel to the focussed plane)
- Later a microscope-video of an object being moved in x-, y-, and z-direction (i.e. including depth changes)
To Do
- Generate stitched image from the input-video (linear complexity desirable) without feedback from microscope-drive
- Cross-compare images to avoid a drift of the estimated shift
- Later provide extended depth of field by maximising a focus measure.
See Also
External Links
Contact Jan Wedekind for a project in this area.
Automated photo stitching
 |
 |
 |
 |
 |
 |
|
|
Premise
- A set of images taken with the same camera-settings (aperture, exposure time, focal length) and center of projection but different viewing directions (mainly yaw, pitch)
- Manually selected correspondences.
To Do
<math> \lambda\,\begin{pmatrix}m^\prime_{1}\\m^\prime_{2}\\f\end{pmatrix}= \begin{pmatrix}h_{11}&h_{12}&h_{13}\\h_{21}&h_{22}&h_{23}\\h_{31}&h_{32}&h_{33}\end{pmatrix}\, \begin{pmatrix}m_{1}\\m_{2}\\f\end{pmatrix} </math>
- Improve the correspondences using 2D cross-correlation (roll-angle will assumed to be low).
- (Re)implement the panotools-solver preferably in Ruby. Use given correspondences to optimize camera-parameters: rotations, common focal length.
- Adopt existing method for removing fringes (transitions/blending)
Optional:
- Automatically find correspondences
- Optimize more camera-parameters (distortion,...)
Also see panorama viewer.
Contact Jan Wedekind for a project in this area.
External Links
- Photo stitching software PTGUI
- Rodriguez Matrix
- Panorama Tools software suite (mirror of Helmut Dersch's former web-page)
- Hugin
- Depth of focus for panorama tools
- Autostitch (commercial)
Physics Engine
Contact Jan Wedekind for a project in this area.
Premise
A physics engines is useful for simulating robots and testing computer vision algorithms. The Open Dynamics Engine (implemented in C/C++) was used in two projects already.
To Do
The Open Dynamics Engine is not numerically stable. An investigation into numerical algorithms for simulating multiple rigid bodies is required. The rigid bodies can be connected by joints, which are limiting their degrees of freedom.
See Also
External Links
RANSAC
Random sample consensus is a method for object recognition. This project is about recognising macroscopic rigid objects (e.g. household and office articles like cups, stapler, ...
To Do
- Select point-features and a suitable similarity measure
- Implement RANSAC algorithm and apply to at least 3 degrees-of-freedom problem.
- Demonstrate algorithm on real object.
Optional
- Extend RANSAC implementation to more degrees-of-freedom problem (maybe use line- and point-features)
HDR imaging
 |
 |
|
Merging pre-aligned 8-bit colour photos into a High dynamic range image will be of high importance as long as low-cost HDR-cameras are not available for the consumer market.
Premise
- A pre-aligned exposure-series of pictures
To Do
- Detect over- and undersaturated colour-components
- Estimate camera sensivity
- Merge images into a single HDR image
Optional:
- Map to reasonably realistic-looking low-range image
