[GRASS-user] DEM creation from aerial photography

[Tim Bowden]

On Tue, 2013-09-10 at 11:11 +0200, Benjamin Ducke wrote:
> One of the handiest solutions for extracting 3D data
> using SfM is this:
> 
> http://homes.cs.washington.edu/~ccwu/vsfm/
> 
> It includes georeferencing and can handle the coordinate
> ranges that occur in geographical data. It is free
> to use for non-commercial applications but unfortunately
> not open source.

Most of the underlying tools to that are open source, and they're what
I'm using (opensift, bundler_sfm, cmvs/pmvs etc). I've had some limited
success so far (not tried aerial imagery), but I don't yet properly
understand the work flow or all the options at various stages of the
process. I also don't understand yet how to measure the 'quality' of the
output (rather crucial). Much more experimentation needed...

> 
> Anyhow, SfM works best when there is a large series
> of images with a lot of overlap and small angular
> displacement between consecutive images. Traditional aerial
> imagery will not always give good results. There can also
> be trouble with the simple camera model that some SfM
> tools (including VSFM) use.

Yes, as I understand it, each point should be in /at least/ 3 images,
preferably with about 15° difference in viewing angles between camera
positions.

> 
> An open source solution that is optimised for remote
> surveying applications is currently lacking, AFAIK.
> 

I think most of the pieces are there assuming you have the cpu & ram to
cope with the processing load but they're certainly not integrated into
a nice GUI. Also being able to 'rubber sheet' the final point cloud to
ground coordinates is not something I've found yet; Doing a '7
parameter' conversion is there, but that's not quite the same.

Idle thought: Is this something that could be worked into GRASS?

Tim Bowden