[OSM-dev] "IQ Routes"-style averages from actual users driving the road

Aaron Whitehouse lists at whitehouse.org.nz
Tue Feb 1 10:27:33 GMT 2011


Hi Eric,

Sorry for the delay in replying.

On 18/01/11 10:44, Eric Marsden wrote:
>   For this to work, it really needs to be integrated into a navigation
>   application for mobile devices, as already implemented by Tomtom and
>   Google for instance. This helps fix the "vehicle category" problem
>   that you mention. Collecting data from mobile devices without running
>   down the battery too much (low frequency logging, simplifying paths on
>   the device to avoid sending overly verbose tracks) and giving people
>   an incentive to contribute is a project in itself. Perhaps there could
>   be a collaboration with mobile-app consumers of OSM data such as
>   Skobbler?

I certainly see what you are saying, but to me a lot of those things
seem like future features, rather than essential things to get this off
the ground. As one example, I currently have a TomTom XXL and use
TTTracklog to collect .gpx tracks for OSM. To my mind, these would be
ideal for traffic analysis, as they are only ever recorded in the car
and I assume the GPS is pretty good. The TomTom is always plugged in and
I'm not that worried about battery life on it. I upload them manually
from home, so the size of the tracks isn't an issue. A number of tools
already produce .gpx files, so it would be great to be able to use those
where the developer has no intention to add specific support.

If tracks need to be less verbose for traffic analysis, the most
sensible approach in my (not-very-informed) opinion would seem to be to
take the overly-detailed track and process it server-side, so that the
intelligence only had to be written once. With OSM clients available for
so many platforms and under both Free and non-Free licences, I would
have thought that as much should be server-side as possible, to lower
the barriers to integration/contribution. This would have the added
bonus that my overly-verbose track could be used to trace roads as well.

In an ideal world, I completely agree that the implementations would
balance things like battery life, upload size, privacy (maybe not
recording points within a changing random radius of "Home" etc) and
convenience. I also agree that more than half the battle is getting
people to contribute. An alliance with someone like Skobbler (or a Free
program like OSMAnd) would be one approach. Another is to let
enthusiasts like me submit more manually prepared data until the system
is useful enough to encourage adoption.

>   Furthermore, the data could be used for other purposes, such as
>   automatically identifying probable missing roads in OSM, probable
>   missing oneway tags, missing turn restrictions, etc.

This is probably the case even without any traffic analysis stuff, isn't
it? Something like Skobbler does your routing for you and presumably
*could* realise that more than half of the people told to turn right at
an intersection did not do so, so an OSM Bug for someone to check for a
"no right turn" is added. Or that people drive somewhere that OSM
doesn't know is a road (probably also possible from any .gpx track
flagged "vehicle").

One thing that could be added if Traffic Analysis was implemented would
be something that flagged when more than x% of people drove more than y
km/h over the speed limit (or vice versa), to catch places where the
speed limit is likely wrong. That is something that the TomTom is pretty
bad at for me.

Again, these would all be brilliant, but would likely be limited to just
Skobbler, just OSMAnd, just Android or whoever implemented that feature
-- unless you have implementations upload itinerary files along with gpx
tracks, but that has its own set of privacy and copyright problems.

>   I mentored Lukas' project in the last Google Summer of Code, but I
>   don't currently have sufficient time/energy to contribute to the
>   data collection aspect of this project. If anyone feels enthusiastic,
>   go ahead! 

Thank you so much for your contribution so far! What is there looks like
it is 90% of the way. I really hope somebody else can pick up the torch
and turn it into something really useful!

Hopefully this overly-long email added something to the conversation!

Regards,

Aaron



More information about the dev mailing list