NOTE: The side-numbers used below are the ones used by your pdf reader,
not the ones given on the slides.
This seminar was held autumn 2009


Slide 1: 

Note that I cut away the original first slide since it only contained
some informations on how I wanted to present the seminar. That's why
it's beginning with slide 2 now.


Slide 2:

Hollywood conducted a study on how many went to the 2D version of a
movie and how many went to its 3D counterpart, if both versions were
showed in the same town (alt. area of town for large metropolitan
areas). Although ticket prices for 3D were larger and there were
normally less 3D than 2D cinemas in the areas, the majority went to
the 3D version of the movie. Unfortunately I couldn't find a link to
the original study, only some articles in online magazines.
Two things should be considered here however:

1. It was probably not possible to pirate copy the 3D versions of the
film. Therefore only the 2D version might have suffered from file
sharing.

2. This study is aimed at 2D cinemas to persuade them to switch to 3D.
Hollywood wants 3D. Not only do they hope that it might make up
for their losses in recent years, they also use 3D as argument to
sell digital systems to the cinemas. Many cinemas still use analogue 
techniques and don't want to switch since it costs them money. However,
fabrication and distribution of digital movie media is less
expensive than analogue, so Hollywood would benefit from not using
film reels anymore.


Slide 3:

Quality aspects: of course the display systems are constantly
refined. The quality got much better over time, especially eye
strain and artifacts were reduced compared to earlier systems.


Side 6:

Motion parallax is widely considered as one of the most important 3D
cues. Its effect is demonstrated in [1].

Eye fatigue is probably the most important problem of 3D vision
systems. The focus aspect however is something which is only
seldomly taken into account. A more detailed explanation of it
(along with an example) can be found in [2]. This aspect should be
examined further in a study.


Side 7:

see also [3].


Slide 9:

see also [4]; the pictures were taken from there.
I don't trust this study too much, since only few subjects were
involved. Especially the right-eyed vs. left-eyed aspect seems to
have some strange results. But what we can learn here in any case is
that a higher quality observed by one eye can cancel out a lower
quality observed by the other.


Slide 10:

For more usage studies see [5], and maybe [6] (also it is more about
mobile 3d).


Slide 11:

picture taken from [7]. The little portable picture boxes for
children which have 2 eye holes and present a variety of
threedimensional pictures are based on this technique as well.


Slide 12:

picture taken from [8]; it shows a stereoscope from the 1893 World's
Columbian Expedition. Note that the technique itself outdates
photography.


Slide 13:

VFX1: a virtual reality system from the late 90s aimed at private
usage (picture taken from [9]). It wasn't a big success however. See
also [10].
NASA uses VR for different purposes like training and remote control
of robotic equipment. Picture taken from [11], see also [12].


Slide 14:

Picture taken from [13].


Slide 15:

More scientific correct would be of course “wavelength multiplexing”.


Slide 16:

The acer laptop was rumored to be released already this year [14]
(picture taken from there as well). A first look of it can be found
in [15].
Asus seems to offer something similar[16] though I heard it uses
actually polarization (see page 19+).
These are the first laptops following Nvidias specifications, but
not the first ones with 3D displays ever. Others can be found from
e.g. Sharp[17].
Stationary systems using Nvidias technique are also already
available, e.g. from Samsung or Viewsonic. More information can be
found under[18].


Slide 17:

The technique used by SEGA[19] works unfortunately only for old tube
TV sets. Of course, the image and 3D quality of the new laptops /
computer screens is much increased compared to the old Master System.


Slide 18:

After-image: due to switching time of the LEDs.


Slide 19:

This movie caused a 3D craze which wouldn't hold however.
Picture taken from [20].


Slide 20:

The company producing the IMAX 3D technology is RealD[21]. The same
also outfits Sweden's SF cinemas (if they are 3D capable).
For an explanation of light polarization see wikipedia [22].


Slide 21:

The report can be found under [2].


Slide 22:

Picture taken from [23] (which contains a good introduction to the
field). In the left, the parallax barrier works like a pinhole
camera which lets each eye view a different image. In the right, the
same effect is achieved by a concave glass which diffracts the light
rays in different directions.


Slide 23:

Brightness and crosstalk have to be balanced out for a good display.
Since the LEDs emit a cone-formed beam their light  might “leak” to
the sides. Removing this (e.g. by using narrower slits) reduces
crosstalk, but also the brightness.


Slide 24:

Although  it is easily possible to introduce motion parallax, this
is only very seldomly done. There are many papers on introducing
motion parallax by the use of head tracking and others on single
user autostereoscopic displays using head tracking to adjust the
stereo images, but I could only find one paper combining the two
[24], which was also more about data reduction via view selection.


Slide 25:

Picture taken from [25]. Note that it is partly wrong: you will
probably never have a whole view between your eyes with current
systems.
Note also that the different manufacturer handle the definition of
“view” differently. While for some it means only one image, others
use it to describe a pair of images with which a stereoscopic effect
can be achieved. This gets even more confusing with newer displays
(see below).


Slide 26:

Guardbands: to avoid stereo inversion, a “black area” is introduced
between the different views, where no image is projected, which
means of course that the viewer will only received an image at one
eye when he is partly in that area.
Note that most modern displays use a different technique to solve
this problem: they define that there is a distant of 5 to 6 cm
between the views (which corresponds to the average eye distance),
thus allowing each vimage to be either a left or a right image.
Example manufacturer for these displays are Spatial3D[26] (which
somehow seem to be connected to Wazabee[27] and Experience3d[28]),
Zerocreative[25] and Alioscopy[29]. I heart the last one delivers
quite good displays, but have never seen any kind of comparison.
These kind of displays are already used, mainly for advertisements
at e.g. airports or exhibitions. Actually, advertisement is most
likely the “killer applications” for them: the 3d effect is eye
catching, but since it wears out quickly and the displayed content
might not be to interesting,  probably nobody will look long enough
at them to notice their shortcomings.
For a demonstration of such systems see [30], which is further
described in[31].


Slide 27:

Note that the pixels in these displays are distributed among the
different views, i.e. the resolution is divided by the number of the
images.


Slide 28:
see also [32]. Google has a book preview which actually contains the
part on super multiview.
I have never heard about anyone who actually built such a display.
The few I found in the literature are more holographic displays.
However, there might be a different solution for this problem. Setred
[33] developed a new multiview autostereoscopic display where
they use a kind of slit mask that is moving horizontally. At any
time, the pixels behind the slit can only be seen on a certain
position is space (a little bit like the parallax barrier mentioned
earlier.) Thus it is possible to use time multiplexing of different
views as well, thus decreasing the resolution reduction of multiview
autostereoscopic displays.


Slide 29:

Voxel is the threedimensionel equivalent to the pixel. It can be
seen as a pixel with an additional z coordinate.
In swept volume displays, the number of voxels is used as a kind of
a threedimensionel resolution.


Slide 30:

Image taken from [34].
This display from the mid-nineties had actually 800000 voxels on a
91,4 cm big helix and full color. It was developed for the US Navy
to e.g. display the sea ground for navigation of submarines.
See [35] for a demonstration of these kind of displays.


Slide 31:
Perplexa display from Actuality Systems[36], see also [37] (and [38]
for a demonstration; note however that the 3D effect gets lost in
the 2D video). I heard that they have one of these in the lab of CSI
Miami (tv series). I don't know however how realistic it is.


Slide 32:
3D cube, image taken from [37].
I saw equivalent systems which were using LEDs instead of a
projector & multiplanar element.
Note that these systems have a noticeable cardboard effect when
there is a big distance between the different sheets. On the other
hand, having a small distance might lead to a bad depth resolution.


Slide 33:
Pictures taken from [39], were this system is described further as
well.


Slide 34:
Unfortunately I didn't found any paper on this, only a magazine
article ([40]; pictures taken from there as well) and a youtube clip
([41]).
The youtube clip has trouble catching the 3D effect however, and it
looks to me that there are less than 100 plasma balls in the air at
the same time as promised in the magazine article. Note that the
sound is typical for this system.
By the time I wrote the presentation I thought it was the nearest
that we come to freestanding holograms. However, I was wrong. See
[42]. Here very fine water drops are used to diffract the image of a
projector. Note that this is not a 3D display however. It could
maybe made to one though, but that would involve to solve a lot of
obstacles.


Slide 35:
For an explanation on how holograms work see wikipedia [43]. The
images were taken from there as well.
Note that holograms are very dense packed with data, that's why the
technique is also used in storage media like e.g. blueray. For our
purpose this is of course a disadvantage. 
Other volumetric displays suffer from similar problems. However, the
data can be compressed quite easily.
More on that later.


Slide 36:
Right picture taken from [44], with modifications.
This technology is probably out of date, especially since there
seems to be no such system that was using different colors
(simultaneously).
Note however that it is not necessary to film the object using the
reference light like it is done in real holograms, at least
theoretically. If the parameters of the lightning used for the
filming are known, and the ones of the display system, it might be
possible to artificially create the correct lightning by digital
signal processing. I couldn't find any papers on this though,
 robably because such systems are dated, as mentioned earlier.


Slide 37:
Picture taken from [45].
This is more how a future holographic system might look like. Using 
acousto-optic modulator (aoms) and optical elements, the lightning
rays are diffracted to create the lightfield of a hologram. Note
that it is a scan based system (like in tube television and older
computer monitors), so it is susceptible to flicker. And it is not
available yet with color (although it is pointed out in the paper
that it should be easily possible to introduce that).
However, it is the most state-of-the-art effort in this area that I
could find.


Slide 38:
The systems on the market are of course swept volume displays.


Slide 39:
There are several different predictions on what will happen to 3D,
from pessimistic (the bubble will burst as it always did before) to
optimistic (3D will be in every home in only a few years).
This is the scenario that I myself think is most probable. Glass
based systems will be available very soon, but 3D won't receive
widespread acceptance for home applications until glasses-free
systems will be available that won't suffer from any shortcomings.
As it looks now, this will be either supermultiview or
electroholographic displays.
Both Sony[46] and Philipps[47] announced to produce glasses-based tv
sets, which will be available from next year. Both are using
polarization. Note that Philipps also showed a prototype using
shutter glasses. From a technological viewpoint, I would give the
edge to polarization, since these systems can be designed to be 100%
flicker and (nearly) 100% crosstalk free. Additionally, the glasses
are cheaper. However, the system itself might be cheaper with
shutter glasses, which might be the selling point for customer.
(Since probably only early adopters will buy them anyway the price
might not play such a big role though.)
Philipps also only recently (summer 09) abandoned its
autostereoscopic displays.


Slide 40:
Some definitions for the next slides. 3D can be seen as a mutiview
stream with 2 streams which were both filmed from the same angle and
a position that varies only in its y-part by a certain, constant 
distance.


Slide 41:
Some more definitions. Note that these are only used in this
presentation, but I wanted to be able to separate the actual encoded
video stream in its entirety from the different view streams it may
contain. Such a definition seems to be missing in the literature.


Slide 43:
Picture taken from [48], where this solution is described in further
detail.


Slide 45:
The same picture as on slide 43, but with modifications. This
solution is mentioned in [48] as well as in other papers, but I
couldn't find any results.


Slide 48:
Picture taken from [49].


Slide 49:
Although it was mentioned that additional streams might be added to
solve the occlusion problem in a few papers, I couldn't find any
which actually does that.


Slide 51:
Picture taken from [50]. See also there for a more detailed
description of the algorithm.


Slide 52:
Although this was mentioned in a number of papers, I couldn't find
any which actually describe such a solution and its results in
further details. The main problem is of course to choose which views
needs to be transmitted and which not.


Slide 54:
Picture taken from [51]. There this algorithm is explained in
further detail as well.
It seems to be mainly based on an earlier paper [52] from (nearly)
the same authors.


Slide 56:
Example paper [53] (and [54] from the same authors). It is a pity
that they don't give any numbers on complexity. Since the
performance is otherwise the same it would be interesting to know if
their approach needs less complex hardware / computational power.


Slide 57:
An example hole filling algorithm can be found in [55]. Note that it
uses a depth map which is calculated from the view stream, not given
in advance.


Slide 58:
Picture taken from [23]. 
Ghosts result e.g. from the fact the LEDs can't switch from one
color to the other as fast as necessary. Therefore, a little bit of
one image can “spill” to the other.


Slide 59:

Picture taken from [23].


Slide 60:

Example paper: [56].


Slide 61:
The main idea of rendering for volumetric displays and
electroholographic displays  is to remove voxels which are either
transparent and colorless or occluded. 
With electroholographic displays, only the fringes (lightrays with a
given starting point and direction) should be rendered which
actually can be seen. These can than be further subsampled.
It is also possible to use some basic fringes and to get more
different fringes by combining these basic fringes.
For more on compression and rendering for electroholographic
displays see [57].