OpenStreetMap
A couple of ideas on how to improve quality/speed of buildings import (on the example of Fremont, CA)
Posted by Yury Yatsynovich on 23 July 2020 in English.1) Combine Fremont buildings layer (pretty detailed, nicely aligned) with MS buildings (poor details, but no substantial biases) … for a centroid of each F-building, find the offset to a corresponding centroid of an MS-building … look at the distribution of such offsets for the nearest 50-100 F-buildings and take the median as an estimate of the offset by which a given F-building should be moved before importing it into OSM
2) For each building compute distribution of angles at nodes, split all buildings into two groups: (a) building with all angles ~90 degrees – these building can be orthogonalized with “Q” in JOSM (or in QGIS) before uploading to OSM (b) buildings that have any angle that is significantly different from 90 degrees (define via hypothesis testing?) – these should be imported without othrogonalization
Discussion
Comment from impiaaa on 23 July 2020 at 17:00
Regarding #1, MS buildings are generated from aerial imagery, which is often misaligned from ground truth. I might trust the municipal data more in this regard.
#2 sounds like a good idea, as long as the detection method doesn’t catch shallow, almost-90° angles that are part of curves. Some municipal data also might have already done a similar orthogonalization pass.
Comment from Yury Yatsynovich on 24 July 2020 at 03:35
For #1, in general I agree with you, but in this case (Fremont CA) when opening both layers (MS and Fremont municipal layer) in JOSM, I can see that MS is aligned fine with Bing imagery (which, in turn, is aligned fine with available GPS tracks), while the municipal layer is slightly shifted – and the shift is different for different blocks. Recently I draw buildings in Berkeley, CA and encountered the same issue – municipal layer was with slight shifts, while MS on average was well centered (though, of poorer details and sometimes slightly rotated relative to the image). Anyways, I’ll try #1 to see if it helps to decrease the shift – if the results are good enough, I’ll post them here:)