[visionlist] [cvnet] Trying to demonstrate stereoscopic vision remotely

Barton Anderson barton.anderson at sydney.edu.au
Sun Mar 22 20:30:51 -04 2020

Hmmm, well I don’t think this is right Ben.  Let’s take the case of two vertical lines side by side.  The left one is nearer than the far one and is the one that is fixated, so it has not vertical or horizontal disparities.  The right line will be slightly closer to the right eye, and therefore generate slight vertical disparities at its top  and bottom (larger in the  right eye). This is unaffected by the convergence angle; the vertical disparities are a function of differences in distance (because the eyes are spherical).  Now consider what happens with two cameras with a flat projection plane.  The right line will  project to the temporal side of the right image, which has been rotated forward; and to the ‘nasal’ side of the left eye, which has been rotated away.  So for flat planes, the vertical disparities will be in the same direction, but the  vertical  disparities  in the flat planes (cameras) will be much larger.  You can actually experience this in pictures where the cameras are rotated; the  regions away from the point of convergence are hard to fuse.

From: Ben Backus <ben.backus at gmail.com>
Date: Monday, 23 March 2020 at 10:25 am
To: "James P. Herman" <hermanj at gmail.com>
Cc: Michael A Crognale <mcrognale at unr.edu>, Barton Anderson <barton.anderson at sydney.edu.au>, David Peterzell <davidpeterzell at me.com>, cvnet <cvnet at mail.ewind.com>, Christopher Tyler <cwtyler2020 at gmail.com>, "visionlist at visionscience.com" <visionlist at visionscience.com>, David Peterzell <dpeterzell at berkeley.edu>
Subject: Re: [cvnet] [visionlist] Trying to demonstrate stereoscopic vision remotely

Thanks James for that.

You know, to be honest I don't think this matters for an introductory lecture on stereo. Stereo effects are pretty robust. There's no harm free-fusing pictures taken with toe-in cameras. You'll still get a nice stereo effect. Better to get students trying things, right?

This is probably not appropriate for most students, but technically, you *can* use toe-in cameras to perfect effect, if the stereo images are displayed in a haploscope, with the arms set to the same convergence angle as the amount of toe-in. In that case the optic arrays would be preserved, so both the perspective and the vergence demand would be correct. That could be the situation Mike C had in mind when he asked the question (either that, or else convergent free fusion of the two images), whereas I was describing what happens when images are displayed in a standard stereoscope or movie theater.

On Sun, Mar 22, 2020, 2:36 PM James P. Herman <hermanj at gmail.com<mailto:hermanj at gmail.com>> wrote:
Ben --

My apologies but I found your answer(s) to Michael Crognale's question difficult to follow, and I was also perplexed by the notion that translation and not translation+rotation is the way to go for DIY stereoscopy!

I found a website that describes the problematic (but not impossible to contend with) distortions of perspective induced by using "toe-in" (relative camera rotation in addition to translation) when taking stereoscopic images.

The short answer is that if you rotate the camera in addition to translating it (between taking photos), the change in perspective can result in identical parts of an image having different sizes in the two photos (the dangerous vertical disparity mentioned previously). I must admit it's not clear to me why any movement of the camera wouldn't potentially cause such distortion, but I suspect it's a matter of scale (translation causes little distortion, rotation causes more). In any case, here's the link:



-- jph

James P. Herman, PhD
Laboratory of Sensorimotor Research
National Eye Institute
Building 49 Room 2A50
National Institutes of Health
Bethesda, MD 20892-4435
office: (301) 496-9376
mobile: (212) 663-0407
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://visionscience.com/pipermail/visionlist_visionscience.com/attachments/20200323/a0c015d9/attachment.html>

More information about the visionlist mailing list