Machine Learning-based Signal Processing Approaches for Automotive Radar Sensors

Traffic scenarios involving both vehicles and vulnerable road users (VRUs, e.g. pedestrians) pose one of the main challenges on the way to highly automated driving. Autonomous vehicles not only have to detect pedestrians, but they also have to be able to analyze their behaviour and intentions, in order to properly adjust their own behaviour for the sake of VRU safety.

Analysing pedestrians with radar sensors is a promising approach for reliable intention recognition, as radar sensors are very robust with respect to bad weather and environmental conditions. By applying machine learning methods to radar sensor data, various types of pedestrian parameters can be derived, such as their position, orientation, and activity, as well as traffic and communication gestures. A variety of such algorithms has already been successfully developed at the institute. However, as both radar sensors and ground truth systems continuously get better, novel approaches become possible to either enhance or extend existing approaches in order to obtain more and more information about pedestrians. Therefore, the goal of this thesis is to develop techniques for enhanced derivation of pedestrian parameters by means of the sophisticated radar sensors and sensor networks available at the institute. Such methods involve for example neural networks that are trained on our custom radar datasets.

The focus of the thesis can be adjusted according to the preferences of the students, with emphasis e.g. on radar simulations, radar signal processing, or machine learning.

Nicolai Kern, M.Sc.XXXXRaum: 41.1.210Telefon: 0731 50-26430E-Mail
Type of Work

Simulations, Measurements, Development and Verification of Algorithms


Recommended Basic Knowledge

Knowledge in signal processing.

MATLAB or Python programming skills are advantageous.

What else?

Start of the Thesis: as of now