Compass Calibration with two subsequent GPS points?

[Stefan Keller]

Dear all

I recently tested a typical show case of AR showing mountain peaks
(Wikitude, Layar, PearAR).

And I have to say that these applications are not yet really
satisfying at least on smartphones mainly because the compass was
wrong: There's is often an error of about 10 degree - of course after
manual calibration (doing loops with the smartphone) and in a
steel-less environment.

Now I think it could be an obvious idea to calibrate a digital compass
simply based on two subsequent GPS point coordinates which are enough
apart?

Does anybody know of any software based calibration of compasses based
on this idea?

Yours, S.

[Thomas Wrobel]

I had the same idea myself when coding, but it would basically require the person to run/walk forward about 10 meters while keeping their phone pointing the same way. Not particularly practical.

One thing that probably is possible, however, is some sort of solar-based positioning.

If the sun is in the camera's FOV, and the software knows its position and time, it should be possible to determine exact baring.

Obviously, the tricky part is identifying the sun, but it should come out as a disc of fixed width of maximum illumination, with the fixed width varying based on the lens type of the camera.....I think. Angular sensing could also be used to help rule out false-positives. (as the phone should know the suns elevation, based on time alone). 

This is a technique I'm going to look into when I get a chance but I've but very busy lately so not much time for experimenting.

[Mike Liebhold]

This just arrived moments ago:  a related resource from O'Reilly ( I have watched it yet )

Alasdair Allan on Making use of iPhone and iPad Location Sensors
With Location Enabled Sensors and Augmented Reality
By Alasdair Allan

This video guides you through developing applications for the iPhone and iPad platforms that make use of the onboard sensors: the three-axis accelerometer, the magnetometer (digital compass), the gyroscope, the camera and the global positioning system. You'll learn how to make use of these onboard sensors and combine them to build augmented reality applications. This will give you the background to building your own applications independently using the hottest location-aware technology yet for any mobile platforms. Learn more: http://post.oreilly.com/rd/9z1zs1fqjdpj5k3kjdkj4apjgd6dpiv0sdltjcug40o

[Anselm Hook]

why don't you just look at the mountain contours?

[Thomas Wrobel]

Sure, if your near mountain, and you have a known database to compare against :)

I think advanced image recognition against a street-view style database is the "best" solution, but I think looking for the sun (or moon) is more achievable in the short term.

[Stefan Keller]

Hi Thomas

Thanks for the answer (so I now the list is not 'dead').

Nice idea with the sun image detection based calibration. That would
fit with AR since the camera is required. Obviously this only works
daytime and outdoor.

> I had the same idea myself when coding, but it would basically require the person
> to run/walk forward about 10 meters while keeping their phone pointing the same way.

Thats one possibility. I think another way would be to ask the user to
align the phone with "compass north" after walking more than 20 meters
- without pointing the same way and without walking on purpose some
straigt line. But I'm not sure yet.

Yours, S.

2010/10/7 Thomas Wrobel <dark...@gmail.com>:

[Thomas Wrobel]

You mean tell the person to walk 20 meters and *then* point there phone directly in front of them. (ie, "straight on" from the direction they walked?). Yes that would work!

Still be a bit inaccurate, but should be less so then 10 degrees.

Of course, once those new satellites finally get going that 20 meters can go down to a more managerial 5 or so.

(but then, by that time, hopefully image-recognition tech would be advanced enough not to need this).

For a stop-gap solution this does seem pretty simple.