3 ms·
run some machine learning algorithms to identify roads If it was that easy we'd already be doing that. 30cm imagery is still very low resolution for tasks lik
by dagw 10y ago
run some machine learning algorithms to identify roads
If it was that easy we'd already be doing that. 30cm imagery is still very low resolution for tasks like spotting dirt tracks under foliage, plus even the current state of the art in image recognition gives far too many false positives and negatives to be really useful without having someone manually double checking everything.
- Piskvorrr 10y agoFB Engineering is getting there, actually. https://code.facebook.com/posts/1676452492623525/connecting-the-world-with-better-maps/ https://code.facebook.com/posts/1676452492623525/connecting-... - of course, they're currently mapping buildings, which is an order of magnitude simpler task. As far as "someone manually checking" - you'd need that in any case, but it is quicker than someone manually doing all the tracing work.
- maxerickson 10y agoFacebook is working on roads too: https://forum.openstreetmap.org/viewtopic.php?id=57942 https://forum.openstreetmap.org/viewtopic.php?id=57942 They also had an earlier attempt that was pretty messy. The identification of roads works pretty good. Separating roads out into main roads and secondary roads and access roads works less well.