noisy 1-second GPS data

Question

So, I come from a programming background, but not a mathematical one. I have a problem that I'm sure has a great mathematical solution!

so, I have GPS data points for every second of a drive a car makes. Due to accuracy issues, if I capture this GPS data from different devices, the points are slightly different, having a bit of noise.

I use this data to work out speed between points and work out acceleration/breaking from this. The problem is, if the points are slightly different between devices, the events captured can be hugely different and I need consistency between devices.

Is there any kind of smoothing etc anyone can suggest that could help me with this?

Many thanks

Try Kalman filters. It might seem complex, but the web is full of information on this subject. BTW I think that math.SE is much better than stats.SE for this question. — dtldarek, Feb 05 '13 at 13:14
Get several GPS devices, average them. The GPS data sent by the satellites is smeared out on purpose... — vonbrand, Feb 05 '13 at 13:31
@dtldarek We have the full data when we process it, so Kalman isn't as useful. — Michael, Feb 05 '13 at 14:40
@vonbrand at the time of processing there generally won't be more than one device doing the same trip, this is just to ensure data accuracy. — Michael, Feb 05 '13 at 14:41
I voted to close as off-topic with the intention of migrating the question to stats.SE, but this did not happen, I don't know why. — t.b., Feb 09 '13 at 12:31

aiao · Answer 1 · 2013-02-05T16:07:48.610

Depends on the amount of processing time you have.

If time is not an issue .i.e. not online, use particle filter.
otherwise use Kalman Filter

orocos BFL, is a c++ library, which I found really helpful and has nice architecture, and is used in embedded applications.

IMHO, stay away from Bayes++; in my application it outperformed (speed) the BFL, but has one of the nastiest software architectures I have ever seen, where every variable and function is called f or f().!

Matlab has its own Kalman Filter libraries.

I have seen some papers that use online particle filters, but they must have went into alot of trouble to multithread the algorithm and have some really badass hardware.

Also, you could use DGPS, but given that you are finding acceleration from position .i.e. finding second derivative numerically, I don't expect that this will be enough though

EDIT: As I understand from the comments, this is offline processing. I would suggest particle filters in MATLAB. Check this out

noisy 1-second GPS data

1 Answers1