Omniview Cameras with Curved Surface Mirrors
OMNIVIS '00 Proceedings of the IEEE Workshop on Omnidirectional Vision
A Complete Panoramic Vision System, Incorporating Imaging, Ranging, and Three Dimensional Navigation
OMNIVIS '00 Proceedings of the IEEE Workshop on Omnidirectional Vision
Approximating a Single Viewpoint in Panoramic Imaging Devices
OMNIVIS '00 Proceedings of the IEEE Workshop on Omnidirectional Vision
Segmentation, Tracking and Interpretation Using Panoramic Video
OMNIVIS '00 Proceedings of the IEEE Workshop on Omnidirectional Vision
Omni-Directional Vision for Robot Navigation
OMNIVIS '00 Proceedings of the IEEE Workshop on Omnidirectional Vision
ICPR '96 Proceedings of the 1996 International Conference on Pattern Recognition (ICPR '96) Volume I - Volume 7270
A Theory of Catadioptric Image Formation
ICCV '98 Proceedings of the Sixth International Conference on Computer Vision
Camera Models and Fundamental Concepts Used in Geometric Computer Vision
Foundations and Trends® in Computer Graphics and Vision
A computer vision sensor for panoramic depth perception
IbPRIA'05 Proceedings of the Second Iberian conference on Pattern Recognition and Image Analysis - Volume Part I
| Hi-index | 0.00 |
Omni-directional sensors are useful in obtaining a 360°field of view with a single lens camera. Omni-directionalstereo imaging systems were designed in our past researchusing a single camera and a mirror consisting of two concentric,radially symmetric lobes. If the central camera-mirroraxis is vertical, a stereo image containing imageryfrom all azimuth directions from two viewpoints can be capturedwithin one image.A vertically posed catadioptric optical system that employsa mirror with a radial profile other than parabolicor hyperbolic cannot maintain the straightness of any nonverticallines. However other mirror profiles are desirablefor resolution distribution, sensor size and manufacturability.These other mirror shapes, including spherical mirrors,are said to be non-SVP (Single View-Point). Since they lacka virtual perspective point, they pose new feature extractionchallenges.A method for processing the imagery from a panoramicnon-SVP catadioptric stereo sensor to reconstruct a three-dimensionalmodel of horizontal and vertical line featuresis introduced. Horizontal line segments are extracted usingthe Panoramic Hough Transform and vertical line segmentsare recognized as straight radial lines. These segments, andclosed shapes they form are matched between the two lobeviews to locate them in three-dimensions.The practical accuracy obtainable with such a systemwas explored, and some of the issues addressed to makethis sensor work in real 3D reconstructions are described.Triangulating a 3D location requires better calibration informationthan is required for the robust feature extraction.Experimental results with synthetic and real images arepresented to validate this model.