[OSM-talk] When images are not calibrated

[stevea]

In Northern California USA, we've struggled with this (slight misalignment due to numerous imagery layers) for at least 10 to 12 years.  Many OSM editors here have become frustrated with great effort, only to find that "another quadrant" (compared to what they were working on) "jumps around" at a wholly different offset once they reach a particular, not-usually-clear-where boundary (of imaging).  This often exacerbates errors as it attempts to correct them!

I agree with the previous posts that offer good suggestions.  While there was (is?) some effort to make mechanical, perhaps even AI-oriented tools or algorithms to improve this with complex databases of offsets, these seem hopelessly over-engineered and continually fraught with error, much like what humans exacerbate, but on steroids as now Machine Learning is doing it.  Maybe that's the case and these should be abandoned, maybe not, I'm not sure, but I have neither good experiences nor high hopes.  In order of preference, I'd suggest to use:

1)  Something like (if not actually) "Land Registry" boundaries, which are correctly and properly brought into OSM with high resolution and accuracy, especially if they are regarded as "perfect" or "authoritative,"

2)  GPS tracks with more-accurate equipment (yes, Garmin devices are "pretty good," my 15-year-old device often gets down to 3 or even 2 meters with open skies) are a good "alignment base" to use, especially when paired with visual imagery,

3)  Existing roads/tracks, yes, especially when substantial curves are involved, so that at least two dimensions are better characterized,

4)  Strava tracks, as while these are "averaged," when they exist (though there may be much noise to sift through and ignore), the results can often be accurate to within 1 meter or even less.

And yes, it's much more important to know that the post box is "on the southwest corner of the intersection" than to know its exact location to within a meter or two.  I agree that "within 5 to 10 meters can be expected," though 3, 2 or even 1 meters is becoming a more-frequent OSM resolution, especially in areas where many volunteers are frequently mapping and updating.  In Europe, especially Germany, our resolution often improves to the decimeter- or even centimeter-level.

This is and will continue to be a long-term improvement in our map, everybody, likely over the course of decades.  As OSM just turned 17, look for better "calibration" (high resolution / accuracy among many sources) further into our "adulthood," rather than our adolescence.