I was wondering which background image sources are the most accurate. I’ve noticed that Bing and Streetview are aligned with each other quite well. However, the Yahoo! imagery seems to be off by quite a bit (at least where I live it seems to be about 8 metres).
All background images may have some degree of shift from ground truth. The reasons are many and varied: source of the original data, precision of the reprojection (e.g., from OSGB to WGS84), how the tiling software works, … So it is never a good idea to assume that any image is better than about 5m precision (in other words about the same as good GPS traces).
In Nottingham the original Yahoo imagery is about 10 m too far S. Much of the city was originally traced from images and its a hell of a lot of work to realign the data so I only do it for bits where I have decent GPS tracks. OS OpenData StreetView data is closer but in general seems a little bit out compared with GPS traces taken on many different occasions.
Best practice is to switch on the GPS layer and align the background imagery with traces, alternatively align existing features in OSM with the imagery then topological and relative relationships will be retained. Before doing so it is worth examining how many people have mapped in an area (e.g., with ITO’s OSM Mapper), and how much of the mapping was done from ground surveys and how much from tracing. Always believe ground surveys ahead of any imagery, however nice it looks.
A couple of other things: there can be substantial discrepancies between different zoom levels and adjacent tiles within a zoom level (data acquired from different sources & at different times); the available imagery (Yahoo, Bing & even OS) can be out-of-date by as much as 10 years.
You don’t say where you are mapping, but the best way is to get out there & do it rather than rely on imagery.
It looks like in Helsinki (Finland) the imagery from areas which are photographed from a low flying aeroplane is pretty accurate. Accuracy seems to be as good as with a reference imagery with a known location error of +/- 1.5 m (RMSE). That is better than achieveable with cheap GPS units. However, outside that area Bing has high resolution satellite imagery which is not at all as accurate, RMSE is perhaps +/- 15 meters. The very high quality area in Helsinki can be identified with Bing Maps application: If it is possible to tilt the imagery it is taken from an aeroplane. Possibility to tilt does not guarantee the accuracy in any other places in the world, ortorectification may still be inaccurate. Biggest source for error is the Digital Elevation Model that is used for orthorectification and only Bing and image produsers know the real accuracy. However, OSM mappers can compare the imagery with GPS traces.
Do not assume tilt images are the same as the vertical ones. In Maidenhead Berkshire the vertical images are about 10 years old, whereas the tilt images were taken about Summer 2008 (judging by the apparent newness of some fence panels I put in).
My main point is: don’t make assumptions about imagery without verifying on the ground (should have stressed ‘may’). If you are confident that the imagery is well aligned (better than GPS) then go-ahead and use it as is: but there is already quite a bit of evidence that in may areas Bing images can be out by as much as 15m.
Gps traces can however equally be out by something like 5 - 50m. Only once you have a whole bunch of Gps traces and can average them out, do you have a reasonable certainty of the true location. In my experience imagery has mostly been more accurate than my own traces, but yes, you should always be careful and judge what the most accurate source is, as it will be different from case to case