Capture and Processing - 3DTV: Processing and Transmission of 3D Video Signals - page 42

depth image sequences, ranging from 64 204. Thus, it has

always been crucial to apply post-processing on them to obtain higher
resolution depth maps, equal to the resolution of the relevant camera
viewpoint.

× 48 to 204 ×

• Limitation on the quality of the captured image sequences - The noises in the

acquired range images are another source of limitation. Noises in the depth
values still cannot be completely removed after calibration.

Time-of-Flight range cameras have widely replaced 3D depth acquisition
systems that are based on a camera and lens. Camera-based systems are rather
old-fashioned and work by estimating the depth fields in a scene based on the
variations in the instantaneous focus. Similarly, the other methods used are
depth estimation for motion and from changes in the geometric structures.
The disadvantage of using camera-based systems in computing the depth
information has been the fact that in order to remove the frequently observed
ambiguities, spatial and temporal correspondences have to be thoroughly
worked out over a sequence of captured images. Accordingly this results in
increased computational complexity and processing costs.
On the other hand, laser-based depth acquisition systems have been
able to capture the depth field more accurately and with a higher range
precision than camera-based systems. Nevertheless, they are bulky and
more voluminous than Time-of-Flight cameras and are more difficult to
operate (e.g. they usually need to be mounted on a unit that can be adjusted
using tilt and pan movements, in order to maximize the field of view for a
particular scene).
In the scope of 3DTV and corresponding content acquisition systems,
ToF cameras have usually been used in combination with supplementary
high-resolution 2D cameras, in order to aid real-time high quality multi-view
depth estimation because ToF cameras produce registered dense-depth and
colour intensity image sequences. Thus, after diffusing the low resolution
depth map values to other supplementary cameras (via a camera coordi-
nate transformation using the extrinsic information of both cameras) and
upscaling the transformed depth pixels intensities using the appropriate
image processing techniques (i.e. exploiting the inter-camera structural cor-
respondences), higher resolution and higher precision multi-view depth map
sequences can be generated. Interested readers are referred to [2], to find
out more about a practical application of such a hybrid model in estimating
dense multi-view depth maps.

2.2.2 StereoscopicVideoCapture

The stereoscopic video can render the left and right views for both eyes to
produce a 3D impression. The binocular disparity, which represents the dis-
similarity in views due to the relative location of each eye, is exploited most,
to help the human visual system to perceive the depth. Other physiological