[OSM-dev] Import a huge amount of GPS data

[Serge Wroclawski]

On Thu, May 23, 2013 at 4:04 PM, Jingmin Chen <jingmin.chen at epfl.ch> wrote:
> Dear all,
>
> Thank you very much for the active discussion. Again, uploading the raw data
> is legally impossible. But contributing processed information will be very
> interesting as OSM data in China is really bad.

I think Tom's concern needs further addressing. We need to understand
how the data can not be used, but derivative data would be free of
this restriction.  This would  be needed before anything else.

> To what I understand, there
> are two possibilities that are technically practical, and I would like to
> have more advices from you on which methods/tools to use.

> 1. Aggregate GPS points to generate centerlines of the roads. Serge
> mentioned TileMill?

My suggestion is that because the data collected is at low interval,
and (according to you) high density that the simplest thing to do
would be to simply render the points as points. From there, visually,
you can try to determine the street geometry *manually*.

This won't be perfect, or even close to perfect. If you do this in a
place with tall buildings, your data will be very noisy. And with such
a low frequency of collection, you're going to have a hard time unless
there is a great deal of data to aggregate over.

> 2. Intelligently infer the road type according to the GPS speed.

Yes, this would be nice to do, but my concern here is:

1. You'd have a hard time doing so with a frequency of 30 seconds. If
your data were 1, 2, or even 5 second intervals, this seems feasible.

2. You'd have a hard time doing so in a place with tall buildings.

3. My solutions are generally "Let's do something we can do in an
afternoon", where as this would seem to require some real thinking and
math, but still ultimately require manual intervention.


The important thing, really, is this legal concern.

- Serge