Dienstag, 30. Januar 2024

Robustness of Multi-Modal 3D Object Detectors

Topic and Goal of the Thesis

3D object detection refers to the task of detecting and localizing objects in a three-dimensional environment, typically using data from multiple sensors such as cameras, LiDARs, and radar. This is a challenging problem that has many applications, including robotics, autonomous vehicles, and augmented reality.

Multi-modal 3D object detection models for have demonstrated state-of-the-art performance on benchmarks like nuScenes. However, their reliance on densely sampled LiDAR point clouds and meticulously calibrated sensor arrays poses challenges for real-world applications. Issues such as sensor misalignment, miscalibration, and disparate sampling frequencies lead to spatial and temporal misalignment in data from LiDAR and cameras. Additionally, the integrity of LiDAR and camera data is often compromised by adverse environmental conditions such as inclement weather, leading to occlusions and noise interference.

A thesis on this topic could explore the robustness and brittleness of state-of-the-art multi-modal 3D object detectors, including the development of novel architectures and training methods, the evaluation of their performance on public benchmarks and datasets, and the comparison of their performance to other object detection approaches. The thesis should study the different modalities, such as camera and LiDAR, and how they affect the performance of 3D object detectors.

Working Points

  • Literature research on multi-modal 3D object detection
  • Literature research on DNN model robustness
  • Evaluation of existing multi-modal models on available datasets
  • Introduction of robustness increasing training mechanism or model designs
  • Training of existing models on publicly available datasets on our compute cluster
  • Evaluate the robustness of the multi-modal 3D object detection method with suitable metrics

Requirements

  • Python / Machine Learning / Computer Vision
  • Enthusiasm for Machine Learning

What we offer

  • private Nvidia A100 compute cluster with ssh access
  • Existing learning framework for machine learning applications

Hinweis: Bitte kurzen Lebenslauf und eine Notenübersicht anhängen.

Kontakt

Till Beemelmanns M.Sc.
+49 241 80 26533
E-Mail

Art der Arbeit

Masterarbeit

Beginn

Earliest possible date

Vorkenntnisse

Python, Machine Learning

Sprache

Deutsch, Englisch

Forschungsbereich

Fahrzeugintelligenz & Automatisiertes Fahren

Adresse

Institut für Kraftfahrzeuge
RWTH Aachen University
Steinbachstraße 7
52074 Aachen · Deutschland

office@ika.rwth-aachen.de
+49 241 80 25600

Wir nutzen Cookies auf unserer Website. Einige von ihnen sind essenziell für den Betrieb der Seite, während andere uns helfen, diese Website und die Nutzererfahrung zu verbessern (Tracking Cookies). Sie können selbst entscheiden, ob Sie die Cookies zulassen möchten. Bitte beachten Sie, dass bei einer Ablehnung womöglich nicht mehr alle Funktionalitäten der Seite zur Verfügung stehen.