Profile
|
M.Sc. Stephanie Käs |
My current research centers on Human Movement Estimation, especially gesture and action recognition using Large Language- and Foundation Models on monocular (fisheye) images and videos. In my previous works, I contributed to data science projects in particle physics and railway engineering. I have gained significant supervision and teaching experience and place strong emphasis on clear science communication, as well as agile project and group management.
!!! Next thesis projects will be offered in November 2025 !!! !!! No more HiWi positions available !!!
Publications
-
Käs, S., Burenko, A., Markert, L. et al.: How do Foundation Models Compare to Skeleton-Based Approaches for Gesture Recognition in Human-Robot Interaction? In Proceedings of the IEEE International Conference on Robot & Human Interactive Communication (2025).
-
Käs, S., Peter, S., Thillmann, H. et al.: Systematic Evaluation of Different Projection Methods for Monocular 3D Human Pose Estimation on Heavily Distorted Fisheye Images. In Proceedings of the IEEE International Conference on Robotics and Automation (2025).
- Dort, K., Bilk, J., Käs, S. et al.: Comparison of supervised and unsupervised anomaly detection in Belle II pixel detector data. Eur. Phys. J. C 82, 587 (2022).
Student Supervision
- A. Weissberg: Fisheye Imagery (Current Student Assistant)
- O. Culha: LLMs for Human Motion Descriptions (Current Student Assistant)
- B. Thal: "Comparison of Person De-Identification Methods in Human Pose Estimation" (B.Sc. Thesis, 05/25-09/25)
- M. Flaig: "Towards Fine-Grained Human Motion Descriptions" (B. Sc. Thesis, 05/25-09/25)
- E. Schönherr: "Evaluating Anatomical Realism in AI-Generated Human Images" (M. Sc. Thesis, 09/25-10/25)
- L. Markert: "Gesture Recognition Using a Video Foundation Model" (B. Sc. Thesis, 09/24-03/25)
- S. Peter: Research internship on Fisheye Human Pose Estimation (11/23-10/24)
- H. Thillmann: "Re-Engineering an Absolute Pose Estimation Architecture: Enabling Extensibility via Modularization" (M. Sc. Thesis, 09/23-10/24)
- A. Burenko: "Stabilization, Tracking, and Gesture Recognition Methods within Skeleton-based Human Pose Estimation Framework" (M. Sc. Thesis, 09/23-09/24)
- V. Hilla: "An Analysis of Error Sources to Improve Temporal Consistency in 3D Human Pose Estimation" (M. Sc. Thesis, 09/23-07/24)
- T. Schellhaas: "Identifizierung von langsamen Pionen durch Support Vector Machines" (B. Sc. Thesis)
- Project Leader: "Stratospheric Balloon Research Project" "StratoGI" (JLU Gießen)
Teaching & Speaking Experience
- Lectureship: Statistics for Geosciences'22/23 (JLU Gießen)
- Exercise Teaching Assistant: (Advanced) Machine Learning, Computer Vision (RWTH Aachen)
- Seminar Teaching Assistant: Historical Milestones of Machine Learning, Current Milestones in Machine Learning and Computer Vision (RWTH Aachen)
- Lab Teaching Assistant: Experimental Physics Lab I-III (JLU Gießen)
- Invited Talk: HASCO Summer School'24 (Uni Göttingen)
- Invited Talk: ErUM-Data-Hub'23/24 (RWTH Aachen, TU Dresden)
- Guest Speaker: JLU Digitaltag'24 (JLU Gießen)
- Invited Talk: Belle II Research Meeting'22 (TUM)
- Public Speaker: Student Hybrid Rocket Team: "HybridLaunch"
Publications
Systematic Evaluation of Different Projection Methods for Monocular 3D Human Pose Estimation on Heavily Distorted Fisheye Images
Authors: Stephanie Käs, Sven Peter, Henrik Thillmann, Anton Burenko, Timm Linder, David Adrian, and Dennis Mack, Bastian Leibe
In this work, we tackle the challenge of 3D human pose estimation in fisheye images, which is crucial for applications in robotics, human-robot interaction, and automotive perception. Fisheye cameras offer a wider field of view, but their distortions make pose estimation difficult. We systematically analyze how different camera models impact prediction accuracy and introduce a strategy to improve pose estimation across diverse viewing conditions.
A key contribution of our work is FISHnCHIPS, a novel dataset featuring 3D human skeleton annotations in fisheye images, including extreme close-ups, ground-mounted cameras, and wide-FOV human poses. To support future research, we will be publicly releasing this dataset.
More details coming soon — stay tuned for the final publication! Looking forward to sharing our findings at ICRA 2025!
