Graphics Reference
In-Depth Information
the responsiveness of the system, and is able to achieve fast restabilization. Normal
moderate speed movement will thereby not be visually noticed by the participants.
6.4.2 Adaptive Nonuniform Plane Distribution
This method of changing the nearest and farthest depth plane is very powerful for
videoconferencingwith one person. There are, however, situations where this method
will not work, for example, when multiple people are standing and walking around.
Therefore, we present an optimization where the distribution of the planes is adapted
to the actual scene content, instead of only the nearest and farthest depth.
A histogram is calculated of the resulting depth map. This histogram guides the
plane distribution for the next temporal frame. This will redistribute computational
power to the more dense regions of the scene, and consequently increase the quality
of the interpolation by reducing mismatches and noise.
When the scene consists of a limited range of depths between
D
min
and
D
max
,
some processing resources are allocated to depth planes where no objects are present.
This can be in between other objects. This is demonstrated in Fig.
6.12
a. Here, a lot of
planes are placed in the scene where no objects are positioned. This wastes resources
and introduces more noise due to mismatches between the cameras. Therefore, we
rearrange the distribution of the depth planes to provide less planes in depth ranges
with less objects, and more, dense planes in scene regions with more objects. We
determine the interest of a depth by analyzing the previous frame in a temporal
sequence. The method works best when the movement of the scene is limited, such
as moving people or scenes with many static objects.
After the interpolation step, we generate the histogram of the depth map using
the well-known occlusion querying method [
15
] on GPU, allowing fast processing.
The histogram can be seen in Fig.
6.12
b. The occurrence of every depth value, as
determined by the depth of the depth planes in the depth map, is counted. The
histogram has discrete depth values between
D
min
and
D
max
, represented by the
Fig. 6.12
a
Uniform plane distribution.
b
Histogram of the depth values
