As part of my mandatory research practice, I designed and implemented the image processing subsystem as well as networked control communication of a visually controlled inverse pendulum. The system uses geometric markers to identify the current pendulum angle. To achieve stable balance, the system requires ultra-low end-to-end delay which is achieved by optimizing the performance of every system component.
During this project, I co-authored a conference paper: Bachhuber, Christoph; Conrady, Simon; Schütz, Michael; Steinbach, Eckehard: A Testbed for Vision-Based Networked Control Systems. 11th International Conference on Computer Vision Systems, 2017 (Link)
Noise Modeling and Synthesis for Motion Picture Production
In my master thesis, I studied film grain and digital camera noise and used a correlation-based model to capture their essential characteristics (grain appearance and strength). This model can be used to calculate a synthesis filter to artificially create authentic noise that can be added to images for creative reasons. After the proof-of-concept implementation was finished and the thesis was written, I implemented the newly developed algorithm as a OpenFX plugin with GPU acceleration (CUDA).
Below you can see a demo of different film grain models:
