Copy the page URI to the clipboard
Bowers, David
(2019).
Abstract
Can you really trust a smartphone Augmented Reality (AR) app to identify navigational landmarks?
AR for navigation depends crucially on three components: position, orientation, and known relative positions of points of interest (landmarks). Mobile AR can be successful for pedestrian tourism, typically in urban environments. For this kind of short-range navigation, it is feasible to correct for inaccuracies in sensors, particularly the compass, by “registering” the camera image against a 3D model of the real world. Without registration, the orientation is derived solely from the internal compass.
For long range navigation, with no nearby landmarks, registration is rarely an option. For example, when approaching a coast in a yacht, the land is too distant, and too small a portion of the camera image, for the computer vision techniques of registration to be effective. Furthermore, surface waves mean that the foreground is moving, as is the whole camera image as the vessel responds to the waves. The lack of registration leads to serious questions about the reliability of mobile AR for navigation purposes, since the internal compass in mobile devices is subject to significant deviation.
I present deviation curves measured for seventeen mobile devices. Deviation curve amplitudes of 5-10 degrees are typical, with comparable linear offsets – sufficient to be unusable, or misleadingly dangerous, for navigation, particularly for identifying features such as leading marks. There are also issues with calibration and with changes in deviation curves over time. Compass errors cannot be corrected simply by “registering” a single known landmark – because the deviation is a time-varying curve, not a fixed offset.
Having demonstrated that current mobile AR techniques should not be trusted for long-range navigation, I speculate on the requirements for interfaces that could harness the potential of AR without succumbing to the limitations inherent in current hardware.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)
