Develop an uncertainty-aware Deep Learning Object Detector and make Autonomous Driving safer!

Institut

Lehrstuhl für Fahrzeugtechnik

Typ

Masterarbeit /

Inhalt

 

Beschreibung

Motivation

In the rapidly evolving field of autonomous driving, the ability to accurately detect and track objects in dynamic environments is critical to ensuring safety and reliability. While advancements in deep learning have significantly improved object detection performance, these systems are not without their limitations. Factors such as sensor noise, environmental variability, and model imperfections introduce inherent uncertainties in detection outputs, which, if unaccounted for, can lead to catastrophic failures in real-world scenarios.

Objective

This thesis focuses on the development and application of uncertainty estimation techniques for object detection in autonomous driving. Specifically, the localization of detected objects (bounding boxes) should be extended with a measure for the prediction’s uncertainty, e.g. it could be formulated as a distribution in space.

Work Packages

• Literature review of uncertainty estimation in/with deep neural networks
• Developing a concept for the evaluation of the uncertainty estimation
• Implementing promising uncertainty estimation techniques into an object detector
• Evaluating the object detector and its uncertainty estimates

Voraussetzungen

A creative mind focused on problem solving, strong interest in machine learning and autonomous driving and an involved working attitude make you the perfect candidate.

Good knowledge of Python is necessary, experience with Pytorch a plus.

Tags

FTM Studienarbeit, FTM AV, FTM AV Perception, FTM Schroeder, FTM Informatik

Möglicher Beginn

sofort

Kontakt

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