[talk-au] Uploading traces (Was; Hi all ...)

[Andy Owen]

On Thu, 2009-06-18 at 15:25 +1000, Ross Scanlon wrote:
> > * improving routing information, by working out average speed on roads
> > (at different times).
> 
> If your connected to the internet that's fine but it's no use for on the
> road re-routing, unless you have all the gps traces downloaded to your
> gps.
> 
> This should be tagged by maxspeed anyway.

Maximum speed is different to expected speed. What I'm talking about is
the ability to work out which road is generally the best at a certain
time. For example, I know if I use the highway at 8, it is clogged with
people going to work, but if I use the alternative route a bit later,
then it is clogged with school parents. This sort of data is mostly
static (when you consider enough dimensions), and unrelated to maximum
speed.

> 
> Some one may have also driven the road really slowly (push bike) and some
> one may have done it at the speed limit.  This would skew any reliability.

Hence the need for lots of data. All the algorithms that do this stuff
are designed knowing that there will be noisy data. Google has their own
version of something like this, except real time. They attach gps
transmitters to some taxis and have sensors on roads, and bring all that
information in to try to work out what the average speed is on different
roads. If a taxi is parked, or a road sensor is busted, it doesn't freak
out and declare that the average speed is 0.

> 
> > * improving height maps, by taking (lots of) samples where altitude
> > information was present.
> 
> Pointless, vertical data is grossly out from a gps you are better off
> using the NASA dem data.

You can still find out useful things from noisy data, as long as there
is enough of it that you can filter out the noise, and since we already
have a starting point for the data, we can be even smarter. Throwing
more information at the problem helps us get a better solution - even if
some of the information is noisy.

> 
> > * automatically guessing the number of lanes on a road, by looking at
> > the variance of traces over sections in each direction.
> 
> Should be tagged anyway (when more than 1) and how do you know it's not an
> accuracy problem.

re accuracy problem: same answer as before.
re "should be tagged anyway": that may be true, but they often aren't.
If an automatic process can fix up most of them, then it saves us time.
And if we revisit the tagging later to make it more expressive (e.g.
saying which specific lanes end), then this will save a lot of time.

> I was going to say look at the sat photo but that dosn't help as its
> covered over with trees.
> 
> We have to trust that osm's are putting in accurate data but from what
> I've seen the data already there is miles better than google maps
> particularly in rural Australia.
> 
Absolutely (and I only checked because it isn't too far out of my way).

But, if someone has the time to do cool things with the gps traces, then
they will be very thankful for any more data. I know of one person who
uses the raw gpx traces from OSM to improve the accuracy of another gps
(it tries to model the walking patterns of a person, so it can predict
how their speed will change as they turn corners and stuff like that). I
don't know a whole lot about it, but this person was begging me for my
gpx traces because he wanted even more data... so obviously there are
some uses for it :)

Just to reiterate - I don't personally care if you do upload or not. But
if you are holding back from uploading because you think it isn't
useful, then I'd disagree.

Andy