Stereo images meant to be seen by binocular human vision are generally taken by two cameras placed parallel to the XY-plane separated by a baseline of +/-65mm.
Given that the vertical disparity can be used to estimate depth just as horizontal disparity, could the images taken by two coaxial cameras (one on top of the other, separated also by 65mm) be used to simulate standard (horizontal) binocular vision? The following article seems to say no but I wanted other opinions... Maybe a reasonable model of the horizontal stereo could be learned from the vertical with ground truth of both types of images?
Hi Bruno,
in principle, the simulation of horizontal binocular vision with the vertical camera set should be possible by reprojecting the stereoscopic image to 3D and subsequently projecting the 3D points to the horizontal stereo camera pair. This will have, however, two drawbacks: Due to the different perspective of the horizontal sensors there won't be a 1:1 matching of the points in the image plane. Secondly, the occlusions from the point-of-view of the vertical camera pair will be lost in the virtual horizontal cameras as well, even though there is no occlusion from their point-of-view. Both effects will result in "holes" in the stereo image. Of course these holes could be filled by interpolation, but this might lead to unnatural effects. If the virtual horizontal cameras are not too far away from the real vertical sensor pair (compared to the distance of the object) the results should be acceptable, however.
Best regards, Ralf
Hi,
I confirm the previous answer. In addition, if you want to "simulate" the horizontal stereo, you need to compute depth for each pixel. If you succeed, then reprojection will be possible (with holes as mentioned above), if not you do not have a chance for convincing visual appearance.
regards
Reinhard
Thank you both for your answers. In the way I see my system, the virtual horizontal cameras would be placed on the same axis as the vertical ones and at the same height.
They used to sell a camera which was actually two cameras, but not vertically but horizontally mounted and activated together. I remember one time when I was getting a new passport or maybe a visa photo, and they used one of those machines. They only needed two photos, but in reality the two photos were a stereo pair.
An alternative for 3d nature photography is even easier - for a majestic view of an architectural or mountain scene, take one photo, then move sideways about a foot and shoot again. You will get a nice stereo pair to use in your stereopticon. The farther away your scene, the wider you should move. You can also exaggerate the effect by moving more. But if you are trying to get a 3-d picture of something very small, you only need move a little bit. This application naturally takes more care as you need to have the two shots exactly at the same height.
You should be able to buy a pair of lenses - then you can make your own steropticon. You can also still buy one, I'm sure.
Martha
P.S. I learned quite a lot on the subject from a visiting philosopher who had made his own very professional stereopticon. I believe his name was Paul Churchill, but there are only about a hundred researchers with that name.
- Badges
- Science topic
- Similar topics
- Papers