Development of a modern Depth Completion Model for the EDGAR Vehicle

Institut

Lehrstuhl für Fahrzeugtechnik

Typ

Semesterarbeit / Masterarbeit /

Inhalt

 

Beschreibung

Introduction

Mono Camera Depth Estimation has made incredible progress in recent years, however,
modern camera only models are computationally heavy and can't be used in conjunction with
other models. Autonomous Vehicles typically have different kinds of sensors, specifically
LiDAR is offering a consistent depth prior.
However, modern depth completion algorithms are typically being tuned for the Kitti-Dataset
whose LiDAR hardware can't be compared with modern vehicles such as the TUM's
research vehicle EDGAR.
Additionally they also don't efficiently utilize odometery, high resolution RGB as well as
exploring the effect of using prior knowledge like Pointcloud maps. The goal of this work is to
develop a real depth completion model that utilizes the current state of the art of research
models well

Work Packages:
- Literature Research about the current state of the art of depth completion and estimation
models
- Develop a lean base architecture for generating RGB-D images in up to 10ms

- Experiment with utilization of further information (Temporal, Odometry prior, Pointcloud maps)
- In-depth evaluation, comparison and iterative improvement.
- Evaluate against stereo camera output

Recommended Literature:

1. Improving Depth Completion via Depth Feature Upsampling
2. Bilateral Propagation Network for Depth Completion
3. Distilling Monocular Foundation Model for Fine-grained Depth Completion

If you are interested or have any questions, please send me an e-mail (niklas.krauss@tum.de) with your CV and a
current transcript of your records, thank you!

Voraussetzungen

Requirements:
- Programming experience with Python and well versed with Pytorch

- High personal motivation and independent working style.

- Knowledge about robotics and optionally already worked with LiDAR

- Very good language proficiency in German or English

Verwendete Technologien

Depth Completion, Python, Pytorch, Machine Learning, Machine Learning, Autonomous Vehicles

Tags

FTM Studienarbeit, FTM Krauss, FTM AV, FTM AV Safe Operation, FTM Informatik, FTM Teleoperation

Möglicher Beginn

sofort

Kontakt

Niklas Krauß
Raum: 3507
Tel.: +49172 1736882
niklas.krausstum.de

Ausschreibung