From July to December 2025, I worked as a Data and Privacy Engineer (Wissenschaftlicher Mitarbeiter) at the Chair of Mobile Systems at the University of Bamberg. My work focused on the BMBF-funded Explanym project, which tackles a fundamental issue in smart city infrastructure: how to collect high-resolution mobility data to improve urban planning without compromising individual privacy.
The Challenge of Geospatial Privacy
Urban planners rely heavily on movement data to optimize infrastructure like bike lanes and traffic signals. However, raw GPS trajectories are extremely sensitive. They can easily expose home addresses, daily routines, and medical appointments. A system cannot claim to be privacy-aware simply by stripping names from datasets; if intermediate raw data is stored or if geographic points are too specific, privacy is compromised.
My primary responsibility was implementing systemic privacy mechanisms directly into the data pipelines. I applied differential privacy techniques to safely aggregate trajectory data, introducing calibrated mathematical noise to obscure unique movement patterns without destroying the statistical utility of the dataset. I also implemented spatial cloaking algorithms to replace exact GPS coordinates with generalized grid cells. This highlighted a persistent engineering trade-off: over-cloaking the data makes tasks like map-matching impossible, while under-cloaking fails to protect the participants.
Data Quality and Map-Matching
During our physical field studies, we collected over 109,000 GPS trajectory points across 415 kilometers of traffic in Bamberg using OpenBikeSensor units and physiological wearables. Real-world consumer GPS data is notoriously noisy, suffering from signal multipath reflections in urban canyons and severe drift.
To correct this, I built a custom map-matching engine leveraging the open-source Valhalla routing project. By utilizing a Hidden Markov Model approach, the pipeline evaluated candidate road segments to reconstruct continuous, logical paths through the street network rather than simply snapping stray coordinates to the nearest road. I also used Dynamic Time Warping (DTW) to align the unsynchronized data streams originating from the independent bicycle sensors and wrist-worn wearables.
Once the data was aligned and cleaned, I utilized Principal Component Analysis (PCA) and t-SNE to reduce dimensionality and applied Bayesian Networks to study probabilistic dependencies between mobility variables.
Recognition and Outcomes
We established a robust, reproducible data engineering architecture—including a full-stack Flask/Dash visualization portal—that allowed the research team to safely process and analyze crowdsensed telemetry.
This technical foundation culminated in a research paper titled “Oh, Be Safe: A Field Study for Crowdsensed Cycling Safety Maps,” co-authored with Debasree Das, Leonie Ackermann, Katharina Ebner, Oleksandr Huba, and Prof. Dr. Daniela Nicklas. The paper was accepted at the IEEE PerCom 2026 conference in Pisa, Italy, where it received the Best Work-in-Progress Paper Award. View the accepted papers list here (Archived).
