Hacker News new | ask | show | jobs
by coldtea 4693 days ago
>It would use a lot of processing power to examine the camera input and attempt to extrapolate positioning for overlayed elements.

What camera input? Why go all in such a roundabout way, processing the image to determine the location?

What he proposes just needs the GPS/compass (if Google Glass has one, else it's trivial to add one). You get relevant wikipedia articles for your location (including the direction you're looking) using Glass's GPS, and the screen displays them.
1 comments
untog 4693 days ago
Nope. GPS isn't anywhere near accurate enough for that- believe me, I've tried. It's usually accurate to a few meters, which is fine for an overhead map, but would be horribly jarring when used for AR.
link
coldtea 4692 days ago
>GPS isn't anywhere near accurate enough for that

For what? I remind you that we're talking about a Wikipedia filter. Wikipedia entries are so coarse grained with respect to position that you don't even need GPS-level accuracy. (Not to mention it can correct itself with the help of the built-in maps, e.g assume you walk in the sidewalk and not 4 meters across inside the buildings etc).

"That place is the Brooklyn Bridge, that thing down the road is the Katz delicatessen, further down is CBGB, stuff like that".

Where's the need for "more than a few meters accuracy"? If it can show me info of the nearby stuff (5-10-20 meters away) on my Glasses, it's perfectly good.

It doesn't have to super-impose it when I look at the specific building (that would be stupid anyway, cause it would necessitate me to keep looking straight at the building to see the info).

link
themstheones 4693 days ago
You need to use military grade GPS. It's much more accurate.
link
mistercow 4692 days ago
Even so, the latency on the compass and GPS is far too high for tracking convincingly.
link
vonmoltke 4692 days ago
Indeed. Military applications that require the type of precision in orientation measurements and derivatives that are being discussed here use high-rate precision Inertial Navigation Systems. INS measurements are made in a coordinate system fixed to the platform. All geometry calculations happen in this coordinate system. GPS and other sensors are used to translate positions in the platform coordinate system back to a "real world" system like lat-lon or ECEF.

This is actually a problem I have been thinking about a fair bit for the Rift. I'm not sure what the best way is to solve it in this package size. I'm thinking a first pass will need some sort of off-board reference.

link