Chair for Computer Aided Medical Procedures & Augmented Reality
Lehrstuhl für Informatikanwendungen in der Medizin & Augmented Reality

H. Najafi
Fast 3D Object Detection and Pose Estimation for Augmented Reality Systems
Dissertation (PhD thesis), Fakultät für Informatik, Technische Universität München, December 2006 (online version) (bib)

Many problems in computer vision and augmented reality (AR) require the estimation of the pose of objects or mobile users in real-time. While reliable solutions have been proposed for pose estimation given correspondences and feature-based 3D tracking, fast and fully automated initialization for tracking, i.e. estimation of the initial pose is still an open problem. The difficulty stems from the need for fast and robust detection of known objects in the scene. This dissertation presents a fast and automated object detection and pose estimation system capable of working with large amounts of background clutter, severe occlusions, and strong viewpoint and scale changes.

The thesis builds upon existing algorithms introduced in the last few years and develops various novel techniques which significantly improve both time and functional performance for detection and pose estimation. The advances can be summarized as two main contributions.

First, a method for a scalable, statistical and compact representation of the object of interest is introduced based on fusion of 3D geometric and appearance information. This is achieved during an offline learning process by a statistical analysis and evaluation of distinctive features of the target object. This representation is then used at run-time during the matching and pose estimation processes to limit the number of hypothesis by incorporating both photometric and 3D geometric consistency constraints. This allows to reduce the effect of the complexity of the 3D model on the run-time performance and makes the method especially for large environments very powerful.

The second contribution consists of a novel sensor fusion framework for pose estimation based on a coarse to fine strategy capable of incorporating multiple sensors, e.g. mobile and stationary cameras. This relies on a statistical analysis, probabilistic estimation and fusion of the uncertainties of the sensors.

Furthermore, this system is integrated into an AR tracking framework for the initialization of a marker-less real-time tracking system, which proves to be fast and reliable enough for industrial AR applications.

This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each authors copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.

Edit | Attach | Refresh | Diffs | More | Revision r1.11 - 19 Jul 2016 - 16:26 - NassirNavab

Lehrstuhl für Computer Aided Medical Procedures & Augmented Reality    rss.gif