Log in

No account? Create an account
More on GPS. - I know it's wonky and I don't care [entries|archive|friends|userinfo]

[ website | My Website ]
[ userinfo | livejournal userinfo ]
[ archive | journal archive ]

[Links:| Randomness Guide to London | Open Guide to Cambridge | Snake Soup | KakeFlickr ]

More on GPS. [Mar. 3rd, 2007|06:40 pm]

After I asked about GPS devices, the very lovely ewtikins kindly lent me her Garmin eTrex so I could try it out. So today I have been happily trundling about London collecting GPS data. I put some of it on the OpenStreetMap public domain list of Tube stations. But I have some questions! And I'm sure someone here can help.

What I'm wondering about is accuracy.

First of all, I've been marking 3–5 waypoints at each place I want to find data for, moving around slightly (~20–100cm) in between each one, the idea being that I might be able to get a slightly more accurate fix in some spots than in others (and the definition of "outside a specific Tube station/pub/etc" is quite fuzzy anyway). Is this a good strategy?

Secondly, the eTrex gives me lat/long to 5 decimal places (I haven't got the cable, so all I can do is read the figures off the display; I don't know if having the cable would give me more decimal places). I don't know if this is rounded or truncated. What I've been doing with my data is rounding it to 4 d.p. and then taking the mode. Is that the best way to do it? doop suggested I instead take the mean and then round that to 4 d.p. Would that be better? Or am I barking up completely the wrong tree anyway?

Finally, in the UK, a difference of 0.0001 in lat or long corresponds to about 10m (unless I've got my sums wrong). This is roughly the same as the tracking accuracy reported by the eTrex itself (5m is the best I've had; the worst has been about 25m; it's generally about 9m). Given that, does it actually make sense to give more than 4 decimal places anyway?

Any advice welcome.

From: hatter
2007-03-03 08:09 pm (UTC)
IIRC, Garmin format will tell you how accurate it thought it was when you took the waypoint. It sounds pretty sane to oversample like you're doing then apply, well, some sort of maths - an average seems a good start, not enough data to do anything more complex, but if you can get an accuracy stat for waypoints too, then you could be a bit cleverer. IIRC you already did some stuff with regions and circles ?

the hatter
(Reply) (Thread)
[User Picture]From: nou
2007-03-03 11:36 pm (UTC)
I can't seem to get the accuracy data out of it by just looking at the screen and pressing buttons; I guess I need to try harder to get hold of a cable!

BTW I was chatting to Mike W-D at a party earlier this evening about this, and your name came up :)
(Reply) (Parent) (Thread)
[User Picture]From: susannahf
2007-03-03 08:36 pm (UTC)
Disclaimer: this may not be helpful, and may actually be unhelpful. Ignore it if this is the case.

Can you get continuous (or at least frequently sampled) data, while you stay in the right general area or stand still (if you get changing numbers when standing still)? If so, and you want to do something flash, you could try Kalman filtering. Basically, this involves building a statistical model of your position (the mean is where you are most likely to be, and the variance is a measure of how likely it is that you actually are). So with point 1, you say "I am probably here (mean is the measurement), and I am this confident about it (variance from GPS accuracy)". Each subsequent measurement updates the model, so it effectively learns what the mean and variance should be.
It's really quite powerful. But it might be overkill.

I know my explanation's not very good. The maths/implementation is pretty easy once you have the equations, although deriving them from first principles is a bit of a pain. Luckily, you won't have to ;)
(Reply) (Thread)
[User Picture]From: nou
2007-03-03 11:38 pm (UTC)
I was chatting to someone earlier this evening about this, and he said the way Real Surveyors do it is to leave something sampling in the same place for a day or so. So your idea's certainly sound! I can record a track, which is either continuous or frequently sampled, but I'll need the cable to get that data off. So maybe an idea for the future!
(Reply) (Parent) (Thread) (Expand)
[User Picture]From: martling
2007-03-03 09:05 pm (UTC)
Have you tried all the output formats? On my flatmate's Garmin if you switch it to the OS grid it gives six figures for both northings and eastings, which is an 0.1m resolution.
(Reply) (Thread)
[User Picture]From: non_trivial
2007-03-03 09:45 pm (UTC)
0.1m is completely unjustified accuracy for a consumer hand-held unit. Accuracy to within 10m sounds about right for in the city, and 25m isn't surprising. IIRC, getting better accuracy requires special kit and doing funky stuff with signal strengths and satellite path times. That gives you accuracy of about 0.1m, but I think this requires you to be stationary for some time - it's used for surveying.
(Reply) (Parent) (Thread) (Expand)
[User Picture]From: nou
2007-03-03 11:33 pm (UTC)
So it does. But still, including all the figures when I report the data does seem to be claiming more accuracy than I actually have.
(Reply) (Parent) (Thread) (Expand)
[User Picture]From: reddragdiva
2007-03-03 09:57 pm (UTC)
Wikimedia Commons has geographical maps plus the original data.
(Reply) (Thread)
[User Picture]From: nou
2007-03-03 11:17 pm (UTC)
Yep, seen that. The OpenStreetMap people are dubious about its non-copyright nature; see the second link in my post.
(Reply) (Parent) (Thread) (Expand)
[User Picture]From: reddragdiva
2007-03-03 11:23 pm (UTC)
Well, the data was collected by the guys who put it on the page there; if they can't say "this here data what we gathered is freely usable by anyone for any purpose without prior permission, knock yourself out", who could?
(Reply) (Parent) (Thread) (Expand)
[User Picture]From: alex [the.earth.li]
2007-03-04 04:12 pm (UTC)


You'll get about 10m accuracy, as that's all the GPS system promises, especially with the ickle handhelds. You can get top-of-the-range stuff which does better, but that needs to run for an hour or so to collect enough signal.

You might also lose something if you're stuck in a valley between buildings as you may not see many satelites. The best approach there is to take another waypoint at a different time (the satelites have 12 hour periods, I think). My experience is that the position fixes from a single reading are good enough anyway.
(Reply) (Thread)
[User Picture]From: nou
2007-03-08 10:35 pm (UTC)

Re: 10m

Thanks - good idea about taking measurements on more than one occasion.
(Reply) (Parent) (Thread)
[User Picture]From: m31andy
2007-03-12 11:11 am (UTC)
Just popping in to say "hi" and "I hope you don't mind that I've friended you". Not only a black pudding connoisseur but also a knitter - yay!!!
(Reply) (Thread)
[User Picture]From: nou
2007-03-12 02:03 pm (UTC)
I actually had my knitting with me in the pub last week, but didn't get around to getting it out.
(Reply) (Parent) (Thread)
[User Picture]From: m31andy
2007-03-14 01:33 pm (UTC)
Ditto - the guys tend to frown at me when I do...

(But I ignore them!)
(Reply) (Parent) (Thread) (Expand)