Tutorials & Workshops, Jan-Michael Frahm

Jan-Michael Frahm
Research Assistant Professor
Department of Computer Science
University of North Carolina at Chapel Hill

Tel: (919) 962 1703
Fax: (919) 962 1699
E-mail: jmf@cs.unc.edu

Home

Research

Publications

Curriculum

Tutorials & Workshops

Computer Vision on GPUs

(Jan-Michael Frahm, "Marc Pollefeys, Mubarak Shah, this workshop is held at CVPR 2008.

CVIU Special Issue on "Virtual Representations and Modeling of Large-scale environments"

(Guest Editors: Jan-Michael Frahm, "Marc Pollefeys,Frank Dellaert, and Jana Kosecka)

Virtual Representations and Modeling of Large-scale environments (VRML)

(Jan-Michael Frahm, "Marc Pollefeys,Frank Dellaert, and Jana Kosecka, this workshop is held at ICCV 2007 in Brazil.

Computer Vision for Augmented Reality (Jan-Michael Frahm, Raphael Grasset)

The tutorial reviews the algorithms and techniques implemented in AR-Toolkit as the state of the art tool for camera tracking in augmented reality. Afterwards techniques for markerless camera pose estimation in real-time will be introduced. For the markerless tracking the state of the art techniques in computer vision will be explained in detail. The introduced methods include algorithms for detection and tracking of salient image features in the video, the robust estimation of the camera pose from the motion of the salient features. In the final part of the tutorial we will discuss the techniques used to improve the performance of the markerless tracking algorithms to achieve real-time. Throughout the tutorial we will show applications of each of the algorithms to enhance the understanding. These examples are used to build a marker based and markerless augmented reality system. [slides as pdf]

3D Camera Tracking, Reconstruction and View Synthesis at InteractiveFrame Rates (Jan-Michael Frahm, Reinhard Koch, Jan-Friso Evers-Senne)

The goal of the tutorial is to review and explain the techniques required for fast 3D reconstruction from video to an audience with image processing background. The tutorial will introduce the techniques needed for structure from motion from video images. These techniques are detection of salient image features and the tracking of these points throughout several video frames. Employing the multi-view relations of these points by epipolar geometry and the trifocal tensor enables the estimation of the camera poses from the images themselves. Leveraging the known camera poses multi-view stereo can be performed. These data form the basis for geometric 3D scene reconstruction and novel view interpolation in free viewpoint video. The tutorial will specifically focus on techniques to improve the speed of the above algorithms to achieve interactive frame rates for dense 3D reconstruction and view interpolation. This is achieved by adapting camera tracking and stereo computation to commodity graphics hardware and by tailoring the algorithms towards realtime performance. Throughout the tutorial we will show applications of each of the algorithms to enhance the understanding and will discuss design requirements for vision systems. [slides as pdf]

Visual-Geometric 3D-Scene Reconstruction from Uncalibrated Image Sequences (Reinhard Koch, Jan-Michael Frahm)

The goal of the tutorial is to give an introduction to the foundations and state-of-the-art techniques for 3D scene recontruction and scene visualisation from uncalibrated image sequences of a real scene. Uncalibrated visual-geometric scene reconstruction combines techniques from computer vision and computer graphics in a natural way and has proven useful when merging real and virtual worlds together, for example with camera motion tracking and scene structure recovery in Virtualized and Augmented Reality. [slides as pdf]