As part of an archaeological project aimed at developing a 'standard' approach to low-altitude UAV surveying I spent a lovely and very productive couple of hours at Stoney Littleton Long Barrow near Bath. The site is amazing and is one of the finest accessible examples of a Neolithic chambered tombs in the UK, with multiple burial chambers which are open to view.
The actual survey was done with a DIY hexacopter, based around an APM 2.6. I used a Canon point-and-shoot running CHDK and an intervalometer script. The actual survey was done in under an hour.
A resulting 3D photogrametry model, created using Autodesk's free to use 123D Catch can be viewed here.
The main goal was not perfection but coming up with a procedure that most people can follow and which gives pretty good results.
If you'd like to get involved we're inviting feedback on our good practice guide.
Comments
@Jack, I use VisualSFM for my 3D mapping. VisualSFM does estimate the radial distortion of each image (actually I think it's done by PMVS-- VisualSFM is just a gui) and creates undistorted images before doing the dense reconstruction. I have not noticed strong distortion in my 3D landscape models but I'm not sure if it's because of this undistortionstep, the fact that I'm using a near-rectilinear lens to begin with, or if I've just been lucky so far. VisualSFM is excellent, though and I highly recommend it.
@John thanks, the interval was 2secs, though it can be longer if you fly higher and yes the camera has to be vertical @Jack CHDK now shots as quickly as the camera is mechanically capable which is ideal @Aaron thanks, that was the very next point I was going to tackle, though as @Hugues points out, a simple way of sorting the issue is badly needed I'll go look at MICMAC
@Aaron, very interesting article (not the maths though :). this is indeed relatively well known that you get flat surface curvature when doing a 3D reconstruction. I know an expert here in Belgium who uses the (relatively complicated because all in CLI mode) MICMAC opensource software (it is in fact a suite of image processing tools developed by the French national mapping organisation) to do his 3D reconstructions and able, if I understoos right, to correct this kind of curvature.
Do you think the curvature of the ground is correct? If not, there is a known issue that I've encountered which you may need to consider: http://ccwu.me/file/radial.pdf
The CHDK intervalometer was pretty bad, years ago, limited to 5 second intervals even though the cameras could do 0.5 second intervals in continuous mode. CHDK didn't support continuous shooting, of course.
That's fantastic! And 123D Catch did quite a good job of stitching it, too. Thanks for sharing!
Nicely done