Capture and Processing - 3DTV: Processing and Transmission of 3D Video Signals - page 49

STAGE

STAGE

Figure 2.6 Sample one-dimensional multi-camera arrangement drawings
(left: linear arrangement, right: arc arrangement)

perception that just a stereoscopic video. Auto-stereoscopic, multi-view
and more sophisticated light-field and holographic displays, which will be
explained in more detail in Chapter 5, are the main focus of multi-view video
applications.
On the other hand, stereoscopic and/or conventional 2D displays are still
suitable to use with the multi-view video content in the context of free-
viewpoint video applications. In other words, multi-view video is useful in
the sense that the users are free to change their viewpoint in the 3D scene.
Details of the multi-view and multi-view plus depth video formats were
presented in Chapter 1 and an overview was provided in Section 2.1.
A one-dimensional multi-camera arrangement is the most convenient
way of achieving multi-view video, because current stereoscopic and multi-
view display technologies are suitable to provide horizontal parallax, i.e.
they work on the assumption that the head movement is in the horizontal
direction mostly. Thus, it is not necessary to capture the visual information
on the top and the bottom of the 3D scene objects. Two popular one-
dimensional camera arrangement types are the linear camera arrangement
(the cameras' optical axes are parallel) and the arc camera arrangement (the
cameras' optical axes are convergent). They are schematically depicted in
Figure 2.6. The key requirements of a practical multi-view capture system
can be summarized as:

•

The Cameras should be well calibrated - The intrinsic parameters of each
camera should be well computed. The Cameras' focal lengths and aspect
ratios should be identical. Ideally, all the cameras used within the multi-
camera array should be of the same brand and model.

•

The rotation matrices should be balanced - For linear 1D camera arrays, where
the optical axes of all cameras are parallel to each other; the 3

×

3 rotation

matrix of each camera should be identical.

• The convergence points of cameras should coincide as much as possible - For arc-

type multi-camera arrays, one should try to adjust the cameras rotation
angles, so that the optical axes of each camera should intersect as much
at the same point as possible. Note that it is very difficult to do a manual