On Positioning for Augmented Reality Systems

In Augmented Reality (AR), see-through Head Mounted Displays (HMDs) superimpose virtual 3D objects over the real world. They have the potential to enhance a user's perception and his interaction with the real world. However, many AR applications will not be accepted until we can accurately align virtual objects in the real world. One step to achieve better registration is to improve the tracking system. This paper surveys the requirements for and feasibility of a combination of an inertial tracking system and a vision based positioning system implemented on a parallel SIMD linear processor array. 1 Inertial tracking for Augmented Reality Position and Orientation Tracking is used in Virtual and Augmented Reality (VR, AR). With a plethora of different graphics applications that depend on motiontracking technology, a wide range of interesting motion-tracking solutions have been invented [1],[2]. Many applications only require motion over a small region, and these tracking approaches are usable, although there are still difficulties with interference, line-of-sight, jitter and latency. A modern six-degree of freedom sensor is the IS600[3]. Its inertial sensor simultaneously measures 9 physical properties: angular rates, linear accelerations, and magnetic field components along all 3 axes. Micro vibrating elements are used to measure angular rates and linear accelerations. The angular rate signals are integrated in a direct manner to obtain orientation. The linear acceleration signals are double integrated to track changes in the position. However, this leads to an unacceptable rate of positional drift and must be corrected frequently by some external source. Hence the system uses an acoustic time-of-flight ranging system to prevent position and orientation drift. The accelerations are also fed into an error estimator to help cancel pitch and roll drift. The yaw drift is corrected by the acoustic range measurements simultaneously with the position drift. This leads to a: Max. Angular Rate: 1200 /sec [3] Angular Accuracy: 1.3 RMS [4] Translation Accuracy: 3.6 RMS(mm)[4] Prediction: 0-50ms [3] Orientation Update Rate: up to 500Hz[3] Position Update Rate: up to 150Hz [3] The question is whether such a tracker is feasible for our Retinal Scanning Display for Augmented Reality [8]. The accuracy needed for registration of the real and virtual images on such a see-through headset involves two aspects, spatial accuracy and latency. The user of a Virtual Reality system will tolerate, and possibly adapt to errors between his perceived motion and the visual result. However, with an Augmented Reality system the registration is within the visual field of the user and hence the visual effect may be more dramatic. For spatial accuracy we use the fact that the central fovea of a human eye has a resolution of about 0.5 minute of arc [7] and that it is capable of differentiating alternating brightness bands of one minute of arc. For latency we can reason that at a moderate head rotation rate of 50 degrees per second the system lag must be 10 ms to keep angular errors below 0.5 degrees. The IS-600 system is measured as having about 1 RMS static orientation accuracy and a 3 RMS dynamic accuracy. We must conclude that although suitable for applications in virtual reality, this accuracy is probably inadequate for AR tracking. To illustrate this we map this error onto the 2D-image domain of our see-through headset. Considering fx the focal length of a video camera (in pixels), Lx be the horizontal image resolution and x the Field-Of-View of the camera, then the ratio of image pixel motion to the rotation angels (in pixels/degree) is: