If we use only one camera, the robot can only recognize
two-dimensional features, not the depth of an object. To overcome
this limitation, binocular camera vision can be adopted
for a robotic vision system. By using inverse projection mapping,
which identifies three-dimensional geometric information
about an object from two-dimensional image coordinates and
the mapping between the camera and image frames, the three
geometric components of the object can be computed, including
the depth of the object. An epipolar line is desirable to
save time determining corresponding points between images
from the two cameras. However, an epipolar line search yields
undesirable noise signals due to incomplete camera setups or a
camera selection based on CMOS. As an alternative, we set the
two cameras parallel to each other so that the distance to
the object Z can be calculated based on simple trigonometric
properties between the object and the camera coordinates
(see Fig. 10):