From: dstamp@watserv1.waterloo.edu (Dave Stampe-Psy+Eng) Subject: Re: Integrated laser arrays for Eyephones Date: Mon, 4 Nov 1991 20:59:50 GMT Message-ID: <1991Nov4.205950.9427@watserv1.waterloo.edu> Organization: University of Waterloo brucec@phoebus.labs.tek.com (Bruce Cohen) writes: >> The shift-register that you described is indeed a neat trick, but remember, >> for decent resolution eyephones we've got a LOT more pixels than any >> display yet made. And besides, you still have to have the frame buffer >> somewhere so > >I don't agree with your statement about the required number of pixels. >First, the requirement for high-fidelity images is not more than about >2000-2500 lines in the horizontal; this was achieved in very high-cost >CRT displays years ago, and building framebuffers of this size is just a >question of whether you can afford it (if you have to ask, you can't). > >Also eyephones have a major advantage over fixed screen displays: they >move with the eye, so they can be designed with more resolution at the ^^^^^^^^^^^^^^^^^ >center than the edges to match the variance in resolution in the eye. >Of course, you can't be as parsimonious with resolution as you could if >you quarantee that the fovea of the eye was always aimed at the center >of the display, but there are still some gains to be made there. I'm not aware of any current research into eye-tracking displays: who's doing this? Head mounted displays, to my mind DO NOT qualify as a good way to reduce pixel counts, even with high-resolution areas (see below). Here is some numbers on the advantages of eye-tracking displays over the standard head-mounted eyephones: For full visual acuity in a display, we need very fine pixelation: 120 pixels per degree of visual angle. However, we can assume less acuity: I have about 20/10 vision with correction, and I rarely miss full acuity. This requires 60 pixels per degree, or 3600 pixels per square degree: abuot 32 million pixels for a 110x80 degree display PER EYE. This is an order of magnitude greater than the best current military aircraft simulators. Some reduction in pixel count can be achived with head-mounted displays with a smaller high-resolution area. However, during normal vision the eye turns about +/- 30 degrees, so we'd need at least a 60x60 degree hires area, which is not a substantial reduction in display pixel count. Looking at the acuity of the human eye, we find that the equivalent pixel density varies as the logarithm of the eccentricity of visual angle from the optical center of the eye, from 120 pixels per degree at the center (fovea) to less than 1 pixel per degree at 40 degrees and greater. If this is integrated, we get a total pixel count of 50000-70000 pixels!!! Therefore, if we can make a display with graded resolution matching the acuity of the human eye, and changing position with eye movements' we should be able to DRASTICALLY reduce the number of pixels to be displayed, and speed up rendering proportionately. Because our display pixels do not match the layout of retinal receptors, we should add about 30% to the number of pixels in the display. Also, we need to expand the high-resolution areas radially to allow for slippage between eye position and display position. This translates into 100k-150K pixels per eye: about half of the data required for a standard color TV picture. I have constructed and done research into a number of eye-movement tracking systems, and errors in tracking of less than 1 degree are routinely achieved with a properly designed system. This leaves the problem of slewing display position to match eye movements: peak eye velocities exceed 500 degrees/sec, and final position error should be less than 1 degree. This is not a easy mechanical or optical problem to solve, but I'm looking at it. One problem that needs to be solved is rendering: pixels are no longer rectilinear, so new rendering algorithms are needed (especially anti- aliasing). This is compensated for by the reduced pixel counts, and the lower acuity at the periphery of the displays could be used to lower the number of polygons to be drawn. Alternitively, several rectangular areas could be drawn with different acuities and overlaid, but this is not as efficient as a smmothly-varying acuity system. The only experiments with systems of this type that I am aware of were done by McDonnell Aircraft in 1970-1974. They developed a nonlinear lens that was used on a camera and also used in a projector to present pictures on the inside of a dome-shaped screen. The lenses caused most of a video picture to be devoted to the center 3 degrees of an image, with the rest of the image squeezed into the outer part. The projector lens reversed the distortion. Both the camera and projector were slaved to user's head and eye motions. Their lens has to be one of the weirdest pieces of equipment I have seen. It weighed 20 pounds and was 8" in diameter. This was partially because was designed to military spec: wide temperature range, achromatic, etc. I'm sure we could do better today using holographic lenses, fresnels etc. There is some evidence from their reports that positioning is more important than speed: a 100 mS delay between eye motions and display settling was not seen by most subjects. However, I have strong reservations about the technique used in that experiment, so this may not be a reliable figure. Having worn bad contact lenses, though, I can say that small delays and distortions after eye movement are not very important. Besides, there are a number of potential ways to make small delays invisible to the user. (Saccadic suppression, masking, etc.) In summary, the advantages of eye-tracking displays are real-world virtual resolutions, and reduced rendering requirements. The disavantages are nonrectilinear pixelation, added complexity in the eyetracker, and the mechanical probles of slewing the display and/or high resolution center of the display. Hope this is food for thought and discussion. -------------------------------------------------------------------------- | My life is Hardware, | | | my destiny is Software, | Dave Stampe | | my CPU is Wetware... | | | Anybody got a SDB I can borrow? | dstamp@watserv1.uwaterloo.ca | __________________________________________________________________________