[OSM-talk] [HOT] Why the HOT obsession with low quality buildings in Africa ?

[Oleksiy Muzalyev]

I do not think that in ten years automated image recognition will be 
capable to map buildings correctly. I remember that still in 70s I read 
that computer translation will replace human translators. However, half 
a century later it is still humans who translate texts, though with the 
assistance of specialized computer programs.

Sometimes I myself cannot recognize on a satellite image what kind of 
building it is, - a warehouse, a cowshed, a factory, etc., or how many 
levels it has got. I have to look at its shadow, surroundings, roof 
surface, to make a guess. I wish in future it would be possible to 
switch in an editor from the satellite imagery to the aerial low 
altitude oblique image to see a building from a different angle.

I do some experiments in this field. For example, recently I filmed this 
historical fort 
https://www.openstreetmap.org/relation/5807132#map=18/46.50897/6.07790 
from the air at low altitude from different angles. These aerial images 
can be accessed from the map in two clicks at the fort's commons category:
https://commons.wikimedia.org/wiki/Category:Fort_du_Risoux

If in future the cost of SSDs capacity drops further, perhaps it would 
be possible to store such oblique aerial images at the map (and editor) 
itself as an addition to the major satellite imagery. And then it will 
be obvious and indisputable at what kind of structure we are looking 
vertically from the space. One such 4K aerial image covers at least 4 
hectares (the approximate area of this fort).

With best regards,
Oleksiy


On 7/4/2018 10:16 AM, Jean-Marc Liotier wrote:
> ...
> I am convinced that the immense majority of those buildings will never be
> corrected. In ten years, we can expect massive campaigns of automated
> image recognition to produce new building layers - but even then the
> extensive conflation will be an horribly tedious job.
> ...