IWAR'99 Breakout Session:

Pros and Cons of Various Tracking Techniques

Summary: Eric Foxlin (ericf@isense.com)

In this breakout session we discussed a range of topics central to the use of tracking systems in AR. The questions posed to the group were:

  1. What will be the prevalent approaches to tracking indoors and outdoors, respectively, in 5 years? How can we choose and evaluate possible component technologies to include in a hybrid combination? Will some applications require approaches different than the most prevalent approaches?
  2. What needs to be tracked, relative to what? Is worldframe necessary? Is 6-DOF necessary?
  3. What are the hard problems still to be solved? How long until we can achieve near-perfect registration? Will artificial fiducials be necessary (for vision-based tracking)? Is large-scale auto-mapping ever going to be possible?
  4. What sensor fusion techniques should be further developed? Are there extensions to the standard Extended Kalman Filter (EKF) paradigm that need to be applied in this application? Are there alternatives to the EKF that could be tried?

There was not much controversy over the type of technology that will ultimately track AR displays - no significant alternatives to vision or vision/inertial hybrid trackers were mentioned. Nonetheless, everyone agreed that there is no single tracking technology suitable for all AR applications.

The question about how long it will take until good registration is achievable spawned an extensive discussion of what will drive the development of better tracking systems. Some members of the group felt that only a high-volume consumer application for tracking would cause sufficient investment to develop high-quality tracking at an affordable price. Games were suggested as the most likely consumer application for tracking, but many feared that the tracking performance requirements for most games will be modest, and the resulting products will therefore not be adaptable for use in AR applications.

The question of what needs to be tracked, relative to what, also led to some conversation. Most participants were very focused on solving the 6-DOF motion-tracking problem for head-tracking AR displays. It seems to be assumed that absolute 6-DOF tracking in worldframe is the main goal, and that 2D image registration techniques or object-relative local tracking techniques cannot eliminate this need for the majority of applications. Furthermore, geometric mapping of the environment by the tracking system was not seen as a goal in of itself, but rather as a (possibly) necessary step towards the goal of tracking the user, or maybe a useful by-product.

Time allowed for only a brief discussion of sensor fusion techniques at the end, but many were in agreement that better modeling of human motion is a key endeavor towards improved Kalman filter implementations.