Evaluation Framework for Uncertainty-Aware 3D Object Detection in Autonomous Driving

Institut

Lehrstuhl für Fahrzeugtechnik

Typ

Semesterarbeit /

Inhalt

experimentell / theoretisch /  

Beschreibung

Current state-of-the-art perception systems in autonomous vehicles rely on conventional machine learning techniques to predict the most probable bounding box for detected objects. However, these systems typically ignore the variation in prediction quality, which can fluctuate significantly depending on the scene and change from one timestep to the next.

To address this, we are developing uncertainty-aware 3D object detectors that not only estimate bounding boxes but also quantify spatial uncertainty—and integrate these with object tracking algorithms.

To support efficient development and benchmarking of this new approach, we are looking for a motivated student to help design and implement an evaluation framework for uncertainty-aware 3D object detection and tracking.

Goals:

• Develop visualization tools for input data, outputs, and associated uncertainties
• Perform statistical analysis of localization errors in 3D object detection
• Evaluate the calibration of uncertainty estimates
• Design the framework to be modular and easy to use—supporting plug-and-play integration with various detectors and datasets

This project offers a unique opportunity to contribute to a novel and impactful area of research, as no such evaluation framework currently exists in the field.

Voraussetzungen

Requirements:

• Strong interest in autonomous driving and machine perception
• Enthusiastic and self-driven working attitude
• Programming experience in Python is a plus

Please send your CV a transcript of records and a short motivation (max 5 lines) to:
cornelius.schroedertum.de

Deadline: April 25, 2025

Tags

FTM Studienarbeit, FTM AV, FTM Schroeder, FTM Informatik

Möglicher Beginn

sofort

Kontakt

Cornelius Schröder, M.Sc.
cornelius.schroedertum.de

Ausschreibung