Stabilizing Augmented Reality Overlays
and Self-organization of Ubiquitous Sensors
The traditional method of generating an augmented reality overlay is
to estimate the overlaid object's 3D position and then apply image
projection. An alternative method is to go directly from overlay-to-image
via projective geometry, giving a 3 fold decrease in error. This
method also allows projection using overlays with curved boundaries.
The second part of my talk is based on the question: suppose you randomly
dispersed a few hundred identical camera-on-a-chip cameras. Is it
possible to deduce where the sensors are and how accurately?
These are topics that I have been working on recently.
If there's time, I'll give a short overview of the other activities
in the vision research lab.