Quicktime VR type technologies use panoramic images to create an
3d experience where the viewer can interactively scroll and tilt
a rectilinear view window on a 360 scene. The viewer is apparently
constrained to a fixed central position but is otherwise free to look
in any direction. With multiple panoramas a "node jumping" feature
allows an all-or-nothing transition to another viewpoint at the centre
of another panorama.

Similar panoramic dewarping can be done with VRML viewers but
here positive measures may have to be taken by the scene author to
constrain the viewer to a fixed central position rather than
this user centrality and immobility being a given of the technology.

Panoramic records of real world scenes can be captured with dedicated
scanning cameras, camera arrays or stitched from image sequences.
Various full or partial view representations are used -cylindrical,
equirectangular, 360 circular fisheye, cubic etc. VRML has a special
Background node which uses six cubic images (front, right etc)
to generate a scrollable panorama effect but it can be better
with VRML panoramas to create simple cubic image boxes in the scene
These image boxes produce an identical panning/scrolling experience as
with the Background node but animation positioning, scaling and
multinode continuity effects can be easier to achieve sometimes
with them than with the Background node.

A panorama converted to a cubic representation and mapped to an
image box in a VRML world can have animations superimposed upon it
- including high level character animations - walking, pointing etc.
These character animations can be given a perspectival quality
which can appear to be consistent with that of the scene - ie eg
a character can appear to walk around inside a panorama. If the
character appears to be in front of scene areas which scale-wise
it should be behind the perspective illusion will fail.

Occlusion effects can be implemented seamlessly with VRML panorama
animations through the use of 24bit plus transparency PNG texture-
mapped panels in the scene. Frontal views from the central panorama
viewpoint of scene objects (Panorama Tools allows very rapid, high quality
rendering of such views) can be rendered and converted into PNG textures with
transparent areas with graphic software such as Fireworks. Then with
programs such as Parallelgraphic's Internet Scene Assembler panels
with these image textures can be positioned accurately so they match up
with the background details from which they were derived. The process is akin
to a stage designer standing in the centre of a cyclorama and ordering
assistants about who are holding paintings on glass. These conceptual
paintings are photorealistic and match particular details of the background
scene. Here is an example:

http://www.culture.com.au/virtual/vrml/gpo.htm

With these foreground image panels in place animations can now
appear to go behind scene features. Additionally and independently of such
occlusion effects positioning and orienting a series of panoramic image
boxes correctly in a common VRML world can be used to produce
cinematic-style action-following animation continuity effects.

Here is an example of this kind of multinode animation:

http://www.culture.com.au/virtual/vrml/scene.htm

Consider an initial panoramic image box - it contains say a character
animation and (optionally) foreground image panels for occlusion effects.
The character walks into the scene. Eventually it will walk through
one of the "walls" of the image box and disappear from view. From
a viewpoint outside the box the character would be seen to come through the
wall of the box. Now if another image box is positioned correctly and
this might involve it occupying much of the same space as the first box
and the viewpoint is switched at some point (before the character appears to
walk through the wall of the first box) to the central viewpoint of
the second panorama then at the viewpoint switch the character will be
seen to be contiuously walking - just the viewpoint has switched to a new
location - but the character will appear to be in the appropriate part of the
overall scene from which both panoramas were derived. If the background
detail behind the character at the point of viewpoint switch is recognizably
the same (except seen from a different angle) the effect is cinematic in its
realism. (On my hardware with Cortona VRML viewer there is an instant
scene switch with certain Render modes - R98 (16 and 24 bit) but not
with the scene change with OpenGL. This is annoying
as otherwise the animation is much smoother with OpenGL
rendering.)

Of course- since the two boxes intersect in space - at the moment of
viewpoint  switch the first image box (and any foreground panels it
contains) must be made invisible and the second image box (and its
foreground panels) made visible. Arranging the scale of the image boxes
to match the scale of ones animation models is a matter of trial and
error but It is often simpler I find to make the characters life size and match
the scale of the image boxes (and panels) to that of the characters.
The observer's eye is at the centre of a large box so if the panorama
was taken at the normal height of a couple of metres or so then characters
will have to be walking on a level somewhat below the middle of the box.
Usually you would want the eye level of the characters to match that
of the viewer so adult humanoid characters would be walking apparently
on an invisible surface two metres below the level constituting the visual
horizon of the image box(es).

In any 3d application the same approach would enable the convenient rendering
of similar animation continuity and occlusion effects with photo panorama
backgrounds but the novelty of VRML version of this approach is that with
this the viewer can freely pan the view as the animations and scene transitions
occur (plus all the rest of VRML's web technology features are at the content
developer's disposal). One can see that there can be interesting hybrid
panoramic vr forms - from pre-scripted camera/animation "movies", to
panning/tilting being the main motif with a character/guide an optional
agent like feature, to actual navigable 3d spaces with panoramas
and image panels constituting layered representations in sections of
the scene.

Copyright: Peter Murphy 1999