andygates

Part of the OSM Haiti response has been to pick over fresh satellite images and mark stuff. Half of that is map-tarting, the other half is looking for interesting places: collapsed buildings and bridges, landslides, campsites. This phase is more or less done now, with the relief agencies rolling that data into the other crisis sources and using it as they will.

I've been staring into sat photos (DigitalGlobe and GeoEye have been utter stars in making their data widely) available all week, and wondering if the process could be automated. Collapsed buildings visible from space go into two patterns, it seems: the roof falls in leaving a tic-tac-toe grid of walls, or the whole thing turns to rubble (many buildings have pancaked instead, with the walls failing and the roof falling intact: these aren't easily visible from directly above). Campsites are erratic splats of tent colours in otherwise plain areas. So some rules could be drawn up:

If in a block of edge-detected buildings there is an area of high noise and it's not green, it's probably rubble.
If edge-detection finds a cluster of boxes about a factor or two smaller than their neighbours, it's probably the rooms inside a house.
If in an area of otherwise plain detail there are spots of noise in telltale colours - tarp blue and tent orange, particularly - it's likely to be a campsite. This is especially true if the site is already geotagged as a park, playing field, car-park, etc (ie an open flat space).

It would need higher resolution imagery, but not by much.