Robotic 3D Vision

What it is about

In this module you will learn methods of 3D image processing for robotics. Focus areas are multi-view geometry, visual localization and mapping (SLAM) with monocular, stereo, RGB-D and inertial sensors, as well as 3D object recognition and tracking. In the end you will be able to practically work on basic tasks such as visual odometry, SLAM and 3D object pose estimation and understand the underlying theories.

What you will be able to do

• Understanding of multi-view geometry and non-linear least squares problems
• Applying indirect and direct methods for rigid image registration
• Conducting and evaluating visual odometry (indirect and direct)
• Combining visual and inertial data (visually-inertial odometry)
• Construction and optimization of visual SLAM systems (pose-graph, bundle adjustment)
• Knowledge of different 3D map representations (TSDF, Surfels, Mesh)
• Detection and tracking of 3D objects
• Foundations of non-rigid registration and SLAM methods

What the module consists of

• VorlesungVermittlung und Erklärung der theoretischen Inhalte
• ÜbungVertiefung durch theoretische und praktische Aufgaben in Gruppenarbeit mit Abgabe von Lösungen und Code

Teaching method

• Vorlesung mit Fragen-Antworten zu Beginn (JiTT)Aufgreifen und Klären von Schwierigkeiten der Studierenden
• Übungen in GruppenarbeitEigenständiges Erarbeiten und Anwenden des Vorlesungsstoffs; Abgabe von Lösungen und funktionierendem Code