Page 232 - Innovations in Intelligent Machines

Toward Robot Perception through Omnidirectional Vision 225
1.1 State of the Art

There are many types of omnidirectional vision systems and the most common
ones are based on rotating cameras, fish-eye lenses or mirrors [3, 45, 18]. Baker
and Nayar listed all the mirror and camera setups having a Single View Point
◦
(SVP) [1, 3]. These systems are omnidirectional, have the 360 horizontal
field of view, but do not have constant resolution for the most common scene
surfaces. Mirror shapes for linearly imaging 3D planes, cylinders or spheres
were presented in [32] within a unified approach that encompasses all the
previous constant resolution designs [46, 29, 68] and allowed for new ones.
Calibration methods are available for (i) most (static) SVP omnidirec-
tional setups, even where lenses have radial distortion [59] and (ii) for non-
SVP cameras set-ups, such as those obtained by mounting in a mobile robot
multiple cameras, for example [71]. Given that knowledge of the geometry of
cameras is frequently used in a back-projection form, [80] proposed a gen-
eral calibration method for general cameras (including non-SVP) which gives
the back-projection line (representing a light-ray) associated with each pixel
of the camera. In another vein, precise calibration methods have begun to
be developed for pan-tilt-zoom cameras [75]. These active camera set-ups,
combining pan-tilt-zoom cameras and a convex mirror, when precisely cali-
brated, allow for the building of very high resolution omnidirectional scene
representations and for zooming to improve resolution, which are both useful
characteristics for surveillance tasks. Networking cameras together have also
provided a solution in the surveillance domain. However, they pose new and
complex calibration challenges resulting from the mixture of various camera
types, potentially overlapping fields-of-view, the different requirements of cali-
bration quality and the type of calibration data used (for example, static or
dynamic background) [76].
On a final note, when designing catadioptric systems, care must be taken
to minimize defocus blur and optical aberrations as the spherical aberra-
tion or astigmatism [3, 81]. These phenomena become more severe when
minimising the system size, and therefore it is important to develop opti-
cal designs and digital image processing techniques that counter-balance the
image malformation.
The applications of omnidirectional vision to robotics are vast. Start-
ing with the seminal idea of enhancing the field of view for teleoperation,
current challenges in omnidirectional vision include autonomous and cooper-
ative robot-navigation and reconstruction for human and robot interaction
[27, 35, 47, 61].
Vision based autonomous navigation relies on various types of information,
e.g. scene appearance or geometrical features such as points or lines. When
using point features, current research, which combines simultaneous locali-
zation and map building, obtains robustness by using sequential Monte-Carlo