© CMG Lee • Dec2003
  Home 
 Basic 3D 
 Projection 
 Holographic 
 Optical addressing 
 Flat panel 
 Survey 
 Accommodation 
 Autostereo 
 Perspective 
 Principles 
 Simulation 
 Cam Display 
 References 
  

Simulation of three-dimensional vision

  1. Overview

    Human-machine interfaces are needed for humans to interpret electronic information. There has been a progression in the complexity and speed with which such communication can occur, from teleprinters to high-resolution full-colour displays.

    However, current display technology is still largely two-dimensional (2D). Research has shown that for applications involving spatial presence, productivity and understanding dramatically increases when the third dimension is simulated.

    The author views the progression from 2D to 3D displays as an inevitable progression, as that of lack-and-white to colour television or analogue to digital motion pictures.

  2. Applications

    Certainly the pioneering uses will be in niche applications, such as remote navigation and surgery [Nakajima,2000]. The Merritt Group found that user understanding of a 3D scene is substantially enhanced when it is presented in 3D [Merritt,1997]. Current applications involve wearing stereoscopic glasses, which while providing certain depth cues, does not allow motion parallax unless coupled with position-tracking technology. Furthermore, glasses tend to be cumbersome.

    Computer-aided design and manufacturing also stand to gain from 3D displays, especially when interpreting wireframe diagrams.

    Another early application is entertainment in specialised entertainment centres, such as theme parks where a large audience volume and high ticket prices justify the equipment cost.

    It may be quite a long while before it is widely available for the home user.

  3. Display architectures

    For most purposes, only horizontal parallax is simulated. This means that motion parallax is present when the eye is moved horizontally, but not when it is moved vertically. It is not a problem as human eyes are located side-by-side and humans tend to move around on horizontal surfaces; their height above ground does not change much.

    The following architectures illustrate the diverse approaches to simulating the third dimension. They can also be categorised into spatially-multiplexed, time-multiplexed and hybrid ones, as with colour displays.

    1. Stereoscopic displays

      Stereoscopic displays allow each eye to view a different image. Viewers wear some form of glasses to separate the images for each eye.

      It exhibits stereopsis and yields a satisfactory simulation. Except in the latest head-tracking incarnations, kineopsis is absent, so the viewer observes a skewed imaged when the head is moved.

      Virtual-reality headsets place a small screen, mirror or waveguide in front of each eye, so that each one sees a dedicated image. These tend to be rather heavy and obtrusive, though.

      1. Spatially-multiplexed

        The first 3D display systems were anaglyphs. These used a coloured filter over each eye (usually red and either blue or green) to filter the complementary colours in the view for the other eye. Initially, only grayscale images were available. Although limited colours could be added to selected images, anaglyphs are still rather uncomfortable to view.

        Anaglyph stereoscopic display
        Red/blue anaglyph stereoscopic display

        E.H.Land pioneered using cross-polarising filters for stereoscopic displays [Land,1940]. Views are projected using light linearly polarised in orthogonal directions. The filters only transmit light polarised in the same direction. This setup allows full-colour images but is sensitive to head orientation. When the polarisers are not exactly perpendicular to the light corresponding to the opposing view, some light leaks through, leading to a ghost image.

        Polariser stereoscopic display
        Cross-polariser stereoscopic display

        A development using circular polarisers [Walworth,1984] solves this problem. One polariser is in the form of a clockwise spiral and the other, an anticlockwise spiral.

      2. Time-multiplexed

        Mechanical and later, liquid-crystal shutter glasses alternately block and expose the vision of each eye. They are synchronised with a projector which displays the correct image when the corresponding shutter is open.

        Shutter-glasses stereoscopic display (left-eye exposed)
        Shutter-glasses stereoscopic display (phase 1: left eye exposed)

        Shutter-glasses stereoscopic display (right-eye exposed)
        Shutter-glasses stereoscopic display (phase 2: right eye exposed)

        Liquid-crystal systems synchronised by an infrared transmitter is now used in the design industry and are marketed under names such as Crystal Eyes.

    2. Volumetric displays

      Rotating light array volumetric display
      Rotating light array volumetric display

      An array of point light sources can be physically scanned (usually rotated) through space, and the lights are turned on at the appropriate moment [Endo,2000]. Alternatively, a screen can be scanned, and a synchronised projector illuminates it with the appropriate cross-section. Rotational displays have a singularity (namely the axis of rotation) where it is more difficult to match the display to the rotating parts, but are mechanically more efficient than translational ones.

      Translating screen volumetric display
      Translating screen volumetric display

      The main advantage of volumetric displays is the enormous angle of view (possibly all 360 degrees). Unfortunately, it can only render transparent objects (as back surfaces are not occluded by ones in front) which are smaller or equal in size to itself.

    3. Holographic displays

      Holographic displays potentially offer the highest quality displays. The extremely demanding pixel resolution and data and computational requirements currently limit the scene to small displays with small viewing angles.

      N.S.Marston [Marston,2000] showed that for a holographic display exhibiting only horizontal parallax, of width w, number of lines L, light of wavelength ?, and viewing angle O, number of pixels, Hologram equation. It is found that high-resolution LCD panels provide a mere 4-degree viewing angle.

      Simple holographic display
      Simple holographic display

      The MIT Media Lab’s holovideo display used an acousto-optic modulator with image tiling to provide a 30-degree viewing angle, though real-time pixel data generation is still a challenge.

      MIT holovideo display
      MIT holovideo display

    4. Autostereoscopic displays

      Autostereoscopic displays are related to stereoscopic displays but users need not wear glasses.

      The first varieties employ parallax barriers or lenticular displays, but these are sensitive to viewer’s position. If the user is in the wrong position, the viewer sees a pseudoscopic image where left and right-eye views are exchanged. Moving parallax barriers can also be made either mechanically or from a stack of LCD shutters [Kollin,1988].

      Parallax barrier autostereoscopic display
      Parallax barrier autostereoscopic display, viewed from correct and pseudoscopic positions

      The Cambridge Display and other similar technologies use optics to project different views at different directions. The next section describes the Cambridge Display.

      The Cambridge Display
      Cambridge Displays using a CRT and another using an LCD panel

      It seems possible to directly scan the “screen” with light beams to generate the image [Kajiki,2000] though the image quality may not be as high as that of a projected image.

      The latest autostereoscopic products track the viewer’s eyes and use a tunable parallax barrier to project views in their directions. Current head-tracking devices have trouble tracking more than one viewer. Earlier versions require the viewer to carry or wear a beacon to indicate their location.

  4. Image acquisition and data issues

    Other than computer-generated scenes where arbitrary viewpoints can be rendered in real-time, images of the scene must be captured from multiple viewpoints. This can be achieved using multiple cameras, or one camera moved around for a static scene. When multiple cameras are used, it is difficult to match the views. An alternative way is to use a large lens and shutters to sequentially expose different parts of the lens to the camera [Shestak,1997]. This is in effect the Cambridge Display used for acquisition instead of display.

    Once the views are captured, video transmission is an issue as image data scales with number of views. Fortunately, there is much redundancy between views. Encoding differences between views reduces bandwidth requirements. It was found that reducing the spatial resolution is preferable to MPEG bit-rate reduction (which creates distracting artifacts) [Tam,2000].

    Furthermore, it is possible to capture and transmit a smaller number of views and interpolate between them [Huang,2000], using “view morphing” [Seitz,1996] for instance. The interpolation is not perfect, however. A more computationally intensive approach uses voxels to recover object shapes [Moezzi,1997].


Copyright CMG Lee & ARL Travis, Photonics and Sensors Group, Cambridge University Engineering Department