My Gopro 3+ Black 180 degree stereo rig

Here is a post about a 3 Gopro/Alaudoptical stereo hemispheric video of a very large scale City Model (physical) ...

http://www.reddit.com/r/oculus/comments/2jxa10/a_virtual_experience_of_a_real_city_model/

One of the perennial problems I've found with 180 3d capture with twin 180 fisheyes are stereo distortions (horizontal and vertical stereo misalignments) in very tilted Rift views of the content -- up or down to the corners of the original 180 by 180 coverage.

These problems are worse if you align and calibrate the two views the most intuitive way -- which is to have them identical with a very distant subject. And this can be done in PTGui say by getting the program to find matching features only in very distant areas of the scene (by masking).

And if you have calibrated at least one of the lenses in your stereo pair accurately then the result of such a calibration will have the cameras parallel and having known fovs and distortion factors, and you will know how one camera relates to the other in terms of values of roll, pitch and yaw offsets. And you can render views in different perspectives (equirectangular etc) from pairs of left/right views knowing that distant features at least will be aligned.

But viewing 180 image pairs (aligned like the above method in PTGui) in regular interactive 3d panorama viewer softwares (for 3d TVs say) , and HMD viewer softwares, falls apart in tilted views up or down towards the sides of the 180/180 coverage). One way to mitigate this with many subjects I think is to use "Viewpoint" alignment in PTGui using control points all over the image area.

https://gfycat.com/FaithfulSnappyHare#?format=gif

You see the L/R image differences in the non-central areas of the L/R images are much less in the Viewpoint optimised side of the gif and this corresponds to a much wider, stereo error free, area you can look around in, in a HMD experience (or 3d screen experience) of the panoramic 3d 180 video.

This is a very non-standard use of Viewpoint optimisation and it is using it as a sort of makeshift "rubber sheeting" image alignment tool. Rubber sheeting software is used, for example, to align aerial photography with a map, where you need to get multiple features, not just image corners, in alignment. Viewpoint optimisation assumes all the points in the two views you are aligning are in the same plane, which is not the case at all here, but still the action of the optimisation is to minimise horizontal (and residual vertical disparities) across the image area in a manner that preserves the global geometry of the image. Particularly we want to minimise the horizontal disparities in the extreme non-central areas of the image pair and it does that effectively here.

You see in the Viewpoint optimised side of the gif that the foreground people's feet particularly have much less movement, like they are stuck in place relative to upper parts of their bodies.

So here is the video pair in SBS 180 equirectangular pair (Viewpoint optimised)
https://www.youtube.com/watch?v=GByN_ru0eT4

and you will see if you download the mp4 and view it in LiveViewRift say (type equir, H coverage 180, V coverage 180), stereo ... that it can be viewed in pretty good stereo even towards the corners of the coverage. If I used the standardly PTGui optimized pairs the larger horizontal offsets in the equirectangular originals in those areas would become vertical offsets in very tilted rectilinear corner views and that would prevent stereo impression in those areas -- and create a blurred impression).

http://www.dropbox.com/s/t9acnrimotqk94m/newtownfestequisbs180.mp4 (44 Mb)

I have been getting good stereo alignment by using
- Precalculated lens FoV and distortion values for all the cameras.
- Adding control points all over camera pairs but setting them to horizontal control points,.
- Adding a couple normal control points to the distance to maintain horizontal alignment.
- Use viewpoint correction sparingly as a last step, constraining other settings as much as possible. But this is because I am also trying to stitch multiple cameras together to also make 360° S3D

By constrain other settings I mean. only optimizing every other camera or optimizing only left or right. But never everything at once.

Hi there, great posts, thanks!

I have been trying to make a waterproof gopro 3d rig for the Rift also, tho mine is far more basic than your own!

The vid from about 1min kinda shows what I've done.
https://www.youtube.com/watch?v=vmVUKet_aCI

I only have the DK1 so far and am going to wait for the CV before doing much more with this.

"j1vvy" wrote:

- Use viewpoint correction sparingly as a last step, constraining other settings as much as possible. But this is because I am also trying to stitch multiple cameras together to also make 360° S3D

By constrain other settings I mean. only optimizing every other camera or optimizing only left or right. But never everything at once.

Yes this makes sense for multicamera arrays as if you are shifting say a L camera's pixels to the nexts R's with viewpoint optimisation then you are moving them further away from the next camera to the left's pixels. It is a different rationale with single camera pairs. I am using viewpoint optimisation with twin 180/180 cameras where its action, if the control points are widely distributed, is to shift the pixels together (laterally) whereever it can. And the net effect is minimize pixel disparities particularly in the peripheral zones of the images. Looking into the corners of 180/180 stereo pairs is equivalent to yawing the view as well as tilting. And it is the yawing that produces the vertical stereo disparities when the view is tilted. If there is no yaw (you are looking straight ahead in the Rift) downward and upward views in the Rift look fine. The action of viewpoint optimisation with points all over the image plane is to minimise peripheral disparities (lateral) so with yawed/tilted views they dont become vertical disparities with tilted+ yawed views. Which like I say is equivalent to using viewpoint optimisation as a local warping tool (rubber sheeting) -- not standard usage like you are intending.

btw for folks thinking of modifying gopro lenses this link has a good tip
.. he describes using rubberised clamps as a wrench to remove the lens without damage
http://goprouser.freeforums.org/after-market-lenses-for-gopro-3-t18458.html#p113594

btw ... I have found a couple more interesting looking Chinese megapixel M12 fisheye lenses. I0 megapixel ratings they say. I have ordered a couple of the 1.29mm fisheyes ($99 each for two).

http://cctv-lenses.sell.everychina.com/p-101147349-1-2-5-quot-1-29mm-10-megapixel-fisheye-lens-m12x0-5-mount-f2-5-185degree-6g-1ir.html

... should give almost full circle on 1080p 60fps


this one looks nice too ..

http://cctv-lenses.sell.everychina.com/p-101190610-1-2-quot-1-58mm-10-megapixel-fisheye-lens-m8x0-5-mount-f3-0-192-degrees-8g.html

Here is another guy who is shooting 2d spherical 360s with three Gopros (Black 3s) with 1.41mm fisheyes (probably Alaud Optical fisheyes like I have been using -- since they are the same focal length). He also uses the 2.7K setting

[edit]... yes he is using the AlaudOptical 1.41mm .. from a Japanese 360 video Facebook group .. this has some interesting stuff in it ...
https://www.facebook.com/groups/1512574208962985/permalink/1523866384500434/

https://www.youtube.com/watch?v=GroxXStarMA
https://www.youtube.com/watch?v=xYxBGjOtmZg

if you are keen (1.36Gb!) you can download the video of the wedding from this facebook post
https://www.facebook.com/groups/360.video.professionals/permalink/1557933224436925/

here is the Reddit post where I described my rig for this kind of thing

http://www.reddit.com/r/oculus/comments/2brgx2/spherical_4k_360180_videos_with_only_3_gopro/

... my CCOM 1.29mm fisheyes are on the way ... exciting!

.. they came .. they are great .. sharper and with less flare because of much better multicoating than the Alaud Optical .. full circle on 1440p and what looks like 185 by 150 degrees on 2.7K

.. actually, I just finished testing both lenses -- one lens is great .. the other is defective (lots of flare) .. damaged in shipping maybe .. though signs of impact are minimal

Here are some short test sequences (2.7K on Gopro 3+ Black) with the CCTV 1.29mm 185 degree M12 fisheye.

The video tests are done with the good copy of the lens, which is excellent, contrast, color and detail-wise compared to the (cheaper) Alaud Opticals 1.41mm I have been using (and both are superior to the Sunex 219 I was using before).

It is slightly shorter in focal length compared with the Alaud Optical meaning it slightly wastes the format at 1440p (the 185/185image circle doesnt quite fill the frame), but it makes 2.7K 30 fps and 1080p 60fps formats more usable for VR (maybe 185 by 145 degrees instead of maybe 185 by 115 degrees with the Alaud Optical.

Here is a cropped 1080p section of a clip (showing Gear VR advertising in Sydney city)

http://www.youtube.com/watch?v=5xFRhiGnp2g
http://www.reddit.com/r/oculus/comments/2mv26y/gear_vr_outdoor_advertising_in_sydney_cbd/

And a couple more short clips (full frame 2.7K upload) -- the lizard clip really shows the better contrast and color and the dancer shows how it can handle background flare in a low light situation

https://www.youtube.com/watch?v=hSf-1Jhe8IQ

I found another source of the 10 megapixel 1.29mm fisheye .. hopefully more reliable .. and somewhat cheaper

http://www.aliexpress.com/item/1-29mm-180-degrees-super-wide-angle-lens-fisheye-lens-sport-camera-lens-M12-mount-lens/32224988585.html

btw .. I shouldn't be a rumor monger but I cant resist ... on a forum I frequent a member with some credibility says there will be a multicamera Gopro sync solution made by Gopro ... for what he is calling a Gopro 4 + .. and he has seen material from a large array of these synced with it, he says

Also ... btw .. One way of shooting stereo still 360 panoramas is to have two gopros with 180 fisheyes side by side rotating and this new gadget -- Motrr Galileo is a nice looking motorized programmable rotator for Gopros and other small cameras

http://www.digitaltrends.com/photography/motrr-galileo-motion-dock-adds-support-gopro/

here is a post with a link to a video I made with the CCOM 1.29mm lens -- the post is about stabilisation with electronic+mechanical gimbal stabilisation
http://www.reddit.com/r/oculus/comments/2nfvjc/mechanical_and_electronic_gimbal_stabilization/

Mediavr, what would be the cheapest, best and easiest way to record 3d 360 degree videos for HMD as of today?

- ?x 10 megapixel 1.29mm fisheye (a $79,00)
- ?x Go Pro 3+ (or 4+)
- Dual Hero System
- ...

Mediavr, what would be the cheapest, best and easiest way to record 3d 360 degree videos for HMD as of today?

cheapest, best and easiest are not the same things here ...

cheapest and easiest is I would think a lot (eg. 16) of Gopro 3+ with circular fisheye lenses in pairs with Dual Hero System backs arranged either in a circle or a sphere looking out. In a circle if you are not worried about nadir and zenith 3d perfection or a sphere if you are (harder stitching) -- you would need good PTGui and After Effects or Fusion (free version now!) or Nuke skills

best -- in terms of 3d perfection -- worst in terms of hardware cost and size and weight -- a dense spherical array of very small cameras (50+) all with genlock for synchronization and most with semi-wide angle rectilinear lenses and all with manual exposure syncing.