In the context of data-driven detection and classification methods, realistic datasets are essential for developing and validating robust models. While synthetic data can be used to cover specific edge and corner cases, real measurement data remains indispensable for both training and validation. Therefore, the aim of this work is to create a structured and reproducible radar dataset that captures a variety of scenarios and system configurations.
Objectives of the thesis
The goal of this work is to set up and operate an automotive radar system operating at 77-81 GHz and to adapt an existing interface for integration into a measurement setup. Based on this, a radar dataset will be created and annotated using labeling strategies (e.g., with reference sensors such as LiDAR or cameras, as well as manual labeling). Finally, the dataset will be systematically documented and prepared for further use.
Research Questions
- How do different radar configurations (e.g., bandwidth, chirp parameters, antenna setup) affect the quality and informativeness of the dataset?
- Which labeling strategies provide the most reliable ground truth data?
- To what extent do purely simulated datasets differ from real measurement datasets, and how does this impact their suitability for data-driven detection methods?
- What effects are introduced by micro-Doppler movements that are often neglected or not modeled in synthetic data?
Work Packages
- Setup and commissioning of the radar board at 77 GHz, including familiarization with different radar configurations and their impact on resolution.
- Adaptation of the interface to integrate the radar board into a measurement setup, followed by the execution of initial measurement campaigns.
- Creation of a dataset with varying radar configurations, including labeling and additional documentation (e.g., via camera recordings).
- Development of an initial signal processing and machine learning pipeline to utilize the dataset (e.g., preprocessing, feature extraction, and basic detection/ classification).
Your Profile
- Basic programming skills (Python)
- Basic understanding of radar technology (advantageous, but not required)
Intersted?
If you are interested, please send a short message to:
Jonas Bönsch: jonas.boensch@fau.de
Request now!