Plane Sweeping in Eye-Gaze Corrected, Teleimmersive 3D Videoconferencing - Advances in Embedded Computer Vision - page 133

Fig. 6.3 Concept of stereo vision. A scene is captured using two rectified cameras. Stereo matching
attempts to estimate the apparent movement of the objects across the images. A large apparent
movement (i.e., parallax) corresponds to close objects (a low depth value)
Let's first introduce the high-level concepts of these techniques, in order to better
understand the advantages of our full system prototype in Sect. 6.3 .
6.2.1 Stereo Matching
Stereomatching uses a pair of images to estimate the apparent movement of the pixels
from one image to the next. This apparent movement is more specifically known as
the parallax effect as demonstrated in Fig. 6.3 , where two objects are shown, placed
at different depths in front of a stereo pair of cameras.
When moving from the left to the right camera view, an object undergoes a
displacement—called the disparity—which is inversely proportional to the object's
depth in the scene. Objects in the background (the palm tree) have a smaller disparity
in comparison to objects in the foreground (the blue buddy). The goal of stereomatch-
ing is to compute a dense disparity map by estimating each pixel's displacement.
6.2.2 Plane Sweeping
To conceptually grasp the concept of plane sweeping, let us take a look at Fig. 6.4 .
Here, cameras C 1 to C 3 are real cameras in an arbitrary configuration, whereas camera
C v is the virtual camera view of which we wish to reconstruct the color image and
the scene depth.