[Talk-GB] bing image alignment

OpenStreetmap HADW osmhadw at gmail.com
Fri Sep 13 16:41:04 UTC 2013


On 13 September 2013 14:58, SK53 <sk53.osm at gmail.com> wrote:

Whilst most of the error terms may well, unfortunately, be true...
>
> OSSV scale is 1 pixel / metre, so accuracy is less than that.

That's a common misunderstanding about spatially, or time quantised
data.  If it were really true, the 300m chip length in GPS would limit
GPS accuracy to much worse than 5m.  This is not just true of
photographic imagery.  It is also true of anti-aliased line art, like
StreetView, or even non-anti-aliased material, as long as lines are
not parallel to the primary axes.  Providing you know what has been
reduced to 1m pixels, and it has sufficient high spatial frequency
component, e.g. footpath edges in aerial imagery and the underlying
vector lines in the maps, one can interpolate to rather better than
1m.
>
> So sources of error are:
>
> Feature generalisation in OSSV, noticeable on buildings & roads

These tend to be local distortions.  By choosing features carefully, I
think you can avoid them when aligning aerial imagery with StreetView.

> Re-projection of OSSV tiles using proj4 using OSGB36, errors of +/- 5m

Errors in the OS survey.

> Given an average road is about 7-8 metres across we are less likely to
> notice this with roads anyway.

Which is why the problem only really starts to show now that there is
extensive, armchair, mapping of individual buildings.


> If one looks at GPS traces for the same footpath walked again and again,
> their spread is quite considerable, perhaps as much as 20 m, although an
> average would probably be close to the actual 2 m path.

Many are mapped on a single pass.

>
> So just like any other survey organisation, ultimately we will need our own
> set of convenient reference points.

Unfortunately, StreetView doesn't seem to have OS' trig points!

My feeling is, that, until we can get our own grid of accurately
surveyed reference points, we ought choose one source that is better
than GPS.  As the US probably uses government maps, for this, now that
they are available to us, they would seem to be the best option.



More information about the Talk-GB mailing list