The goal of this research is to quickly create a compact, accurate representation of the path of a robot. Others have developed techniques for building 3D maps, while this work concentrates on building a descriptive image of where the robot has been. This technique which we call \tubular mosaics," provides an orthographic-like view of the hallways the robot has traversed. An omnidirectional camera produces images that can contain a hemispherical eld of view. Mosaics are built from wedges extracted from each frame. We demonstrate the technique using views perpendicular to the camera motion resulting in the upper half of the tubular mosaic (i.e. two walls of a corridor, including the ceiling.) Each wedge is unwarped into a rectangular strip. Strips from consecutive frames of video are matched and incorporated into the mosaics. Others have created orthographic strip mosaics of areas using a camera looking down at an angle from an airplane. Our work is a generalization of this to multiple directions. Three techniques of varying complexity and accuracy are presented. Results will be useful for robot navigation. Other applications include texture mapping and the creation of a compact representation of the video stream captured by the robot.
[1]
S. Nayar.
Omnidirectional Video Camera
,
1997
.
[2]
Terrance E. Boult,et al.
Physical panoramic pyramid and noise sensitivity in pyramids
,
2000,
Proceedings IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2000 (Cat. No.PR00662).
[3]
Hans P. Moravec.
Towards Automatic Visual Obstacle Avoidance
,
1977,
IJCAI.
[4]
Terrance E. Boult,et al.
Frame-rate omnidirectional surveillance and tracking of camouflaged and occluded targets
,
1999,
Proceedings Second IEEE Workshop on Visual Surveillance (VS'99) (Cat. No.98-89223).
[5]
Shmuel Peleg,et al.
Mosaicing with Generalized Strips
,
1998
.
[6]
S. M. Steve.
SUSAN - a new approach to low level image processing
,
1997
.
[7]
Andrew Blake,et al.
Motion Deblurring and Super-resolution from an Image Sequence
,
1996,
ECCV.