Using a GoPro & Photoscan

[William Ringle]

I've just obtained a Hero 4 GoPRo. I've seen a few messages in this forum, but a couple of questions. I've seen that 1.1 has fisheye support but that the user has to specify that a fisheye is being used. Where is that parameter set? Second, can the 12 mb mode be used with any success? I'd really like to get images of some tight interior spaces, so this would be useful. Third, (maybe not necessary) has anyone done a calibration of the Hero 4 lens that they are willing to share? I have trouble getting close enough to the screen with the Agisoft calibration routine to make it work properly.

Thanks!

[William Ringle]

Answered my own question: tools>camera calibration. I still would like to hear the experience of others with GoPros, 7m vs 12m for instance

[mrb]

All the "medium" FOV does is crop the image.  I always use 12mp wide and get perfectly good results.

[nadar]

pro:
- weight, volume, robustness, easy mounting

cons:
- fisheye => not so sharp in the corners
- automatic expoosure => problems for mosaicking
- rolling shetter => unaccurate restitution

[mrb]

- rolling shetter => unaccurate restitution

This is only really a problem if you're going faster than 15m/s.  Shooting stationary or at a reasonable speed, and it's not an issue.  Just something that you have to be aware of if you're using it from an aerial platform.

[Photogrammetryfacts]

Just a tip: You might find a lens profile for the gopro for lightroom. You can then back process the images there and give photoscan the corrected images.

[nadar]

- rolling shetter => unaccurate restitution

This is only really a problem if you're going faster than 15m/s.  Shooting stationary or at a reasonable speed, and it's not an issue.  Just something that you have to be aware of if you're using it from an aerial platform.

Vibrations can also affect images acquired wit a rolling shutter.
I've read somewhere that the actual exposure time of a rolling shutter setup at 1/1000th sec is somewhere about 1/50th.
Each line is aquired at 1/1000th, but the whole image is constructed much slower.
So: rolling shuttre is fine if
1) the speed is not too high (actually, what is important is the object displacement: depends on the camera  translation, but also on the distance to the object: 15 m/s will be very high is you shoot at 10m)
2) you don't have to much vibrations

theses conditions are probably OK for many applications, but seem difficult to meet for aerial photography.

[bigben]

pro:
- weight, volume, robustness, easy mounting

cons:
- fisheye => not so sharp in the corners
- automatic expoosure => problems for mosaicking
- rolling shetter => unaccurate restitution

Ignore rolling shutter issues. I doubt you'd be moving that fast indoors
Automatic exposure can be a pain.  I tweak the exposure/contrast in ACR and convert to tiff...
Fisheye is not as sharp in corners, but in the corners you will also be likely to get higher reconstruction uncertainty and reprojection error, so in the sparse cloud at least these will get filtered out.

The big one that most people seem to forget is that you have 12MP spread across a wide FOV, so the resolution of the final model will be limited by this.  e.g. an iPhone 4s will get you a higher resolution mesh

For interiors I'd add that the GoPro uses ISO to compensate for low light, so the darker it gets, the higher the ISO it uses... thus more noise. You will need to hold the camera very steady in lower light to avoid image blur.  (bear in mind that the GoPro is designed for action video, so blur in low light is not as important as it is with a stills camera)

[gheflorian]

We also have a GoPRO 4 Black Edition and using it with Phantom 2 with PFV and Data Link for way points. I managed to do only 1 field test and this is a screenshot of the model. I have to say that the software really shines at aligning the photos. The light conditions where not OK in that day so don't expect any miracle
Funny that I didn't have to use the tool to mask the legs of the Phantom when they showed in some pictures.

[jtuhtan]

One quick comment when using any fish eye lens:

If the distance between images is too close (< 1m), SfM software may have a hard time calculating the camera positions and orientations correctly. If you are doing any type of low-altitude imagery (< 20m), you probably only need to use every third image or so...

[bigben]

If the distance between images is too close (< 1m), SfM software may have a hard time calculating the camera positions and orientations correctly.

That has not been my experience.The biggest gotcha I've found with close range work using fisheye lenses is moving the camera too far, resulting in inadequate overlap of things at the edges of the image.

This test with a GoPro was entirely <1m https://skfb.ly/ArCN  Holes in the base are due to a combination of exposure and not enough images around the base.

If you are doing any type of low-altitude imagery (< 20m), you probably only need to use every third image or so...

Height, speed and capture rate would be the main factors here so you can't really generalise on this

[MortarArt]

If you're using a GoPro on a Phantom, which seems pretty common these days, keep in mind that you can mount mirrorless dslrs on them too, and get much better results.

[mrb]

true, but at the expense (besides the camera itself!) of processing time (photoscan) and flight time (extra weight).  You'd likely also lose the ability to adjust the tilt of the camera in flight.

Admittedly, I haven't tried putting another camera on my phantom other than a second GoPro to shoot stereo video, so I'm only guessing.

Do you have any examples of models made with a mirrorless on your phantom MortarArt?  I've thought of mounting a Ricoh GR but can't justify the extra $600 for unknown results.

If you're using a GoPro on a Phantom, which seems pretty common these days, keep in mind that you can mount mirrorless dslrs on them too, and get much better results.

[bigben]

I used a GoPro for early testing pthotoscan. For objects I was getting better results with an iPhone 4s. Smaller image, but also smaller fov, so effective a higher pixel density on the subject.

http://potree.org/demo/potree_2014.12.30/examples/matterhorn.html.. Camera was a Canon S110 if they used the standard drones.

[mrb]

I used a GoPro for early testing pthotoscan. For objects I was getting better results with an iPhone 4s. Smaller image, but also smaller fov, so effective a higher pixel density on the subject.

http://potree.org/demo/potree_2014.12.30/examples/matterhorn.html.. Camera was a Canon S110 if they used the standard drones.

I don't think anyone was questioning the fact that GoPros will work with Photoscan, or give good results.  All of the models in this folder were made with a GoPro 3 from an aerial platform and handheld:
https://sketchfab.com/matthewbrennan/folders/99e9a5ac56a7486cac430327eedc17e3

For aerial photomodeling, a GoPro will be better than an iphone for a variety of reasons, and I would argue even so on the ground. 

I'd really like to see a comparison of the same subject photographed with an iphone, a GoPro, a point-and-shoot, and a DSLR, and then all processed in Photoscan. 
maybe I'll get around to that sometime.