<html>
<head>
<meta content="text/html; charset=UTF-8" http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<div class="moz-cite-prefix">Accuracy is indeed a problem.<br>
An early OSM update I made was moving a borderline by 250m and put
a devotion site from an arrondissement to another.<br>
Since then many corrections of more that 5 m, mainly due to user
being unaware of Bing's offset (at close zoom of course, in some
places only, poisoned gift) and some editors.<br>
I just checked a place where I had spotted a Bing offset before.
National aerial photos 2009 and 2012 were offset by 2.8m. Bing was
almost in the middle <br>
Smart phones have a bad accuracy reputation. I have however bought
one based on a "good GPS" user report and, indeed I verified quick
fix and immediate 4m accuracy by cloudy weather in a veranda.
Former one around 10m and couldn't fix in house.<br>
<br>
On 2014-12-23 18:49, Malcolm Herring wrote :<br>
</div>
<blockquote cite="mid:m7c9vj$7hq$1@ger.gmane.org" type="cite">On
23/12/2014 16:57, Tom Pfeifer wrote:
<br>
<blockquote type="cite">The collection of
<br>
traces over a longer time creates a cloud of traces which
<br>
form a Gaussian bell curve, in density, over the ground
truth.
<br>
</blockquote>
<br>
Except that the position of a node in the DB is the last edited
value, not the mean position of all historical values.
<br>
</blockquote>
Yet, I sketched a funny program that analyzes a GPX trace,
determines when the car stops (I doubt it could work for foot),
takes the mean value of the wandering position until the car moves
again, and creates a POI for JOSM to layer.<br>
The results were surprising for an alpha 0 pre-release pure hack.
After rolling up one's sleeves higher, one could imagine processing
several traces, such as ones recorded by buses to determine the
coordinates of the stops. <br>
I had been surprised by how much the GPX position moves about when
the GPS is stopped (compared to when it moves) and I will test that
with my new smartphone when better weather returns.<br>
<br>
Cheers
<br>
<br>
<table>
<tbody>
<tr>
<td>André.</td>
</tr>
</tbody>
</table>
<br>
<br>
</body>
</html>