Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

CMOS induced skew a problem for 3D shooting?

J

JohnF

Well-known member
Joined
Jan 7, 2007
Messages
291
Reaction score
0
Points
0
I've heard that there are concerns that CMOS chip readout method/time induced skew is a potentially serious problem for shooting 3D.

The comments suggested that it would be/is very difficult to correct this problem as the skew would be the different from camera to camera thus inducing serious parallax problems when doing stereoscopic filming. RED was mentioned as being of particular concern.

I would have thought that a good genlock/sync would match chip readout from camera to camera thus any skew would be matched and no parallax problems "just" skew.

Any thoughts?

JohnF
 
They did an episode on Redcentre about this. It is a problem but it can be controlled to some point by obeying the rules about how fast you pan. The ASC book has rules about film cameras that get a "strobe" because you pan too fast. CMOS chips will get a skew. The podcast points out that the issues aren't identical but loosly related.

You will get skew from an object whizzing through your shot. But in theory alot of that might not stand out because it is there and gone so fast and combined with the motion blur it could be partially masked.

Give the podcast a listen, they know a helluva lot more than I do.
 
Hey John,

CMOS/rolling shutter skew problems should not be a problem for shooting regular stereo. The only case that I can think of where this would be major problem is for some of the work we are trying to do which is very different from a regular stereo shoot.

We have been developing software the very cleanly extracts correctly mapped 3d geometries from still and stereo sequences and more besides. For a single pair of stereo cameras skew does not present disturbing parallaxial artifacts providing things are synched up reasonably well.

The problems start to occur if you try to shoot with multiple stereo cameras at different positions to capture for example a 360 degree stereo dynamic capture of a real life subject and you want all the 3d extracted geometries to fit together precisely to form a single continuous 3d model. [Imagine four pairs of cameras arranged N, E, S, W pointing to a central area with a dynamic figure/subject in the centre.].

When you shoot stereo with a single pair of cameras there are many types of data shadows and occlusions that naturally occur, such as the far side of some one’s face or the data shadow created by a ¾ view and some one’s nose (i.e. you can’t see the cheek on the far side of the nose from a single stereo view). Therefore if you want more complete 3d models from stereo you have to use multiple stereo cameras to capture dynamic sequences. The problem then becomes that each set of fragments from each stereo camera position is skewed three dimensionally parallel to the plane of the given stereo camera’s sensor. Therefore a North, East, South and West arrangement of stereo cameras (eight cameras in all) gives rise to 3d model pieces that won’t fit together properly no matter how precise the rigs and calibration you are using, unless the subject is relatively still. However if you have two pairs of stereo cameras in the same plane, (four cameras in all correctly synched) then the 3d fragments will correspondingly fit together in spite of skew and a dynamic subject. This all assumes stationary cameras and a dynamic subject.

As you can see we would love to have a camera that had an electronic global shutter, high frame rate, and high resolution, but no such cameras are possible as yet. So we either have to trade frame rate or image resolution, and in some cases, some basic image quality caused by high res industrial globally shuttered cameras (when/if used). So compromises are necessary and for certain sequences as long as the action is not too rapid, the rolling shutter artifacts in 3d extraction can be compensated to some degree. I know for example also there are some algorithms that have been applied in ariel photogrammetry to compensate for sensor drift during line scan image acquisition.

Are you planning a stereo shoot?
 
Very interesting stuff there Eric.

Yes I'm exploring the possibility of shooting 3D for a project I'm planning.

In answer to Michael's question the shoot has to be done on parallel rigs.

Shooting on a 4K RED solves a lot of logistic and quality issues for this shoot but are there no hd cameras with global shutter?

JohnF
 
In principal, I much prefer parallel rigs from an accuracy point of view. We generally shoot at wider base separations than most conventional cinematic applications as we need to capture fine 3d detail. The mirror boxes can introduce spatial errors and light loss for exacting work, and can be a trifle ungainly (but are never the less a no-brainer for most quick and dirty stereo shoots as they seem fairly versatile). However one of the problems, with for example a RED ONE is that you can’t squash two RED’s together to get an inter-axial distance much less than 10 cm. For well controlled close ups some times you need 3cm to 4 cm or less for the inter-axial distance. Also if the range of parallax from a great range of depth of the scene needs to be controlled, then a smaller inter axial distances are required, (but gives rise to everything looking like cardboard cutouts set at various distances in 3d space). We have been looking at having the cameras face directly into each other and placing a pair of small precision mirrors at 45 degrees (so the line of sight is forward facing). This has the advantage of much less light loss and much less spatial distortion than a “Beam Splitter” type rig, and yet you can still get down to 3cm base separation or inter-axial distance from large camera systems. The cameras move on various axies and slide, looking more like and anti aircraft gun, but at least the system is more precise and well balanced than other methods.

There are many industrial cameras for example from Imperx and JAI/Pulnix that have global shutter, that have HD res. I think for high quality and precision work large sensor size is important and for good stereo, high resolution (i.e. beyond HD) is important too. We have some high precision designs (we do CAD and CAM as well) for these types of cameras, as well as the RED. However, I don’t think there is a broadcast or cinematic quality HD globally shuttered camera yet. I spoke to Jim about this in RECON, and he says it’s certainly possible, but not right now and I don’t think it is something he is interested in right now. There is also the issue that global shutter appears “sterile” as it is literally freeze frame, something that cinematic folks do not like the look of and therefore RED prefers to recreate a more filmic look with rolling shutter (also rolling shutter is easier to process at the needed bandwidth for “4k” than a global shutter).

However for precision VFX, globally shuttered high res camera (I believe is desperately needed). If you find one please let me know!

We have been wanting to shoot event and show jumping horses in action, in stereo, to extract high quality 3d geometry but have also been concerned about synchronization and skew issues as well for more dynamic subjects. I’m not really sure that industrial HD (globaly shuttered cameras) can produce imagery that would even be acceptable for broadcast? (maybe apart from a really expensive solid state cooled sensor system from Cooke that we had looked at.)

John, what kinds of subjects or object sizes are you wanting to shoot?
 
You must log in or register to reply here.
Back
Top