As digital cameras get better (and cheaper) getting metric data from images is ever more accessible. There are lots of whizzy looking models all over the web of everything from your back garden to the pyramids ; so what is the new photogrammetry offering the surveyor today, and why are surveyors so reluctant to trust the camera as a measurement tool?
- By far and away the biggest constraint is processing software that can cope with the classical stereo case. There are 2 reasons for this: the difficulty of getting a digital stereo view at low cost and the sheer processing power needed for processing co-incident images at a good enough resolution for precise work. There are good solutions out there (crystal eyes, eyeArt etc) but they are not cheap. 3DTV is deploying these technologies too, they work well but punish the wallet!
- Object area control is not simple to achieve without using TST, this is not a problem if you have the use of one but for many these are difficult tools to use, even with simple interfaces like TheoLt.
- Good image capture. Despite the astonishing development of digital cameras good photography is not a given! Few surveyors take photography seriously and the principles of balanced exposure are not applied as much as they should.
- Blinded by scanning, many surveyors see photogrammetry as superseded and perceive scanning as ‘easy’ and photogrammetry as ‘difficult’.
Things are changing and new tools developing, research work in the 90s is now bearing fruit as the big players offer us multiple image modelling options with Photosynth and Catch123D. You take your snaps, upload them to the cloud and back comes your model!
At the DIY end of the software spectrum things are improving too…a good demonstrator of how current generation digital photogrammetry can work is Agisoft StereoScan
There is no control data used, no stereo view was required : the model is simply a surface generated by digital image analysis and processing. Close inspection of the exported surface in Meshlab reveals some curious features in the model:
there are bumps in the surface where no bumps should be! From Meshlab I can export the mesh as a DXF into CAD:where I can edit out the rogue points ( it is mostly the canopy of the hornbeam tree) and (thanks to TheoContour) generate a contour plot of the terrain. So it’s quite possible to get from KAP snap to 3D plot for a contour survey just by taking 2 photographs!The contour interval is 4 to the metre indexed at a 1m interval and the scale seems to be pretty close to my memory of the site.
I will need to check the precision and orientation of the results (there seems to be something odd about the Z axis orientation I can’t quite figure out) before declaring this a viable topo survey technique but it’s encouraging to see the surface generation potential from low elevation imagery like this.
The free version of Stereoscan is limited to a single model and doesn’t appear to accept control points for exterior orientation. The full version costs ‘professional’ money (€ 2,660) so I’ll be testing the freebie out pretty thoroughly before I commit but so far, despite the ‘bumpy’ surfaces, it has proved really easy to use.
A comparison with laser scanning may not be appropriate but I can tell you the cost of this data path is very low compared to the scanning equivalent- a photograph from 60m up is a very cost effective 3D capture device!