hi all,
We have been doing some photogrammetry testing for the last couple of weeks. All cameras were hanging in gimbals with wibration damping. I am still not convinced that the compact digital camera on a smaller multirotor is not the best solution;
An octo with Tarot 5D2 and Nikon D800 - big, expensive, knee-shaking in the air, expensive batteries - way too long to recharge. Gimbal working overtime to stabilise the weight. Very slow to take the shot, even with focus fixed and in manual. Delay between firing infrared and photo being taken. Often missed shots. Shots often a bit blurry.
Large quad with Nikon 1 - also delay between APM firing up infra red and shot being taken. Shots less blurry.
Arduphantom with Canon S110 running CHDK intervalometer - very sharp. Never missed a shot.
So the question is; if you can buy a compact digital, a compact DSLR or a full DSLR what is best for photogrammetry? Is the massive cost, effort and stress of a full size DSLR rig worth the 8 extra megapixels?
Really interested in your opinions.
We're using Pix4U for the photogrammetry.
AusDroid
Replies
It depends what you use the data for. sensefly and gatewing demonstrate that you can get very reasonable results with compact cameras. An investment into something much more expensive needs to be weighed against the gains. Also think about the risk of putting all that gear in the air and what that means as risk for a commercial business.
For matching purposes there are often constraints on resolution, because each photo is associated with a feature matrix of a certain size. So even if you could take 30MP photos at intervals that were good, the processing software is probably not going to be able to deal with them. That limit (not for all softwares I suppose) is somewhere around 10-16 MP. This means that if you increase resolution beyond that you're likely going to lower results instead of upping them.
The distortion and sharpness of images are however markedly different (in theory, considering your post). When a lot of distortion occurs you lose resolution, but there's also pixel stretching going on when the photo is undistorted. So a camera without distortion is always better. Chromatic aberration is another issue.
So personally I don't see the point of big DSLR's for photogrammetry considering what you theoretically gain and the good results people are getting now. I think a main decision point between the two is altitude. A multirotor without gimbal for perfect nadir is often unnecessary. You probably get 5-10 deg oblique shots, but these are actually beneficial to get better height accuracy in the final resuts at the cost of slightly worse perfect orthophotos.
Gerard,
Thank you for your reply.
Of course in terms of aerial photography/video, it makes sense to fly bigger cameras with better optics and the ability to get closer with a bigger lens.
In terms of photogrammetry, yeah, I think I have to agree that the compact cameras do a good job. We fly an ebee and the result is OK. The results are stunning if you run additional perpendicular paths so that you cover the area twice.
I just can't get my head around the fact that you get better results with a $200 camera than you do with a $2000 camera.
Has anone out there attempted photogrammetry with a compact DSLR?
Ausdroid
Well, I didn't say that the results of a pocket camera are better than a DSLR, but lifting the DSLR vs a pocket camera brings along significant engineering challenges. So it makes sense to evaluate which results are good enough and at our altitudes of 60-80 meters I think the results are good enough.
DSLR's have less distortion, should in theory allow more light to reach the sensor, but unfortunately due to their weight also have a lot more momentum. It requires bigger motors, more lift per prop, which increases the amount of bending per prop and the probability of being slightly out of balance. All that contributes to a lot more vibration on the model overall.
The difference on photogrammetry is that the precision of DSLR's, *if* their lens is good as a camera is about optics not about sensor size, then you get slightly better accuracy in the 3D model.
Here's one guy using a DSLR: http://www.drmattnolan.org/photography/2013/