VAMPIRE, Visual Active Memory Processes and Interactive Retrieval (IST-2001-34401)

In traditional computer vision systems, processes for model acquisition and object or event recognition are decoupled in time and control. Coupling both kinds of processes poses to significant challenges for computer vision. The model acquisition task has to be performed by active memory processes that have to implement visual learning and categorization capabilities. Usual recognition process must be substituted by active retrieval processes that can reason about the memorized and perceived input data.

The VAMPIRE project aims at the development of an active memory and retrieval system in the context of an Augmented Reality scenario. The user wears a head-mounted camera and display such that the system is able to build up a hierarchically structured memory of what the user sees. Query results are re-visualized in a virtual world or projected into the real world using the head-mounted display.

The MT9 is part of our hybrid (vision-based and inertial) tracking concept which is used to determine head pose for Augmented Reality (AR) in real-time. In contrast to Virtual Reality (VR), which is widely used for simulation, AR enriches the perceived reality by additional information with a representation ranging from video annotation or highlighting to projections of complex 3D objects.

This technique is used as visual aid for medical and military purposes, for entertainment, for assembly processes or for engineering design or - in our case - for interactive teaching of our visual active memory. For this purpose, AR applications must be able to recover the position and orientation of the user's head. This localization task requires high accuracy and low jitter tracking methods performing at real-time. The MT9 is used to improve the performance of the vision-based components especially for fast rotations of the user's head. In addition, it can be employed as backup source of information, if the vision-based components provide no valid pose information (no visible landmarks, etc).