Omnidirectional Vision Frank Dellaert, Georgia Tech CS 3630 Many Slides adapted from Chris Geyer and Emanuele Menegatti See also the Page of Omnidirectional Vision http://www.cis.upenn.edu/~kostas/omni.html Animal Vision - 1 • 2 Millions species • Few hundred distinct organisational plans BUT only two basic types of imaging eyes of wide use: •Single lens camera like eyes •Multi-lens compound eyes From R. Dawkins: Climbing Mount Improbable. Norton, 1996 2 Animal Vision - 2 Q: Why are perspective systems insufficient and why is field of view important? A: Perspective systems are one imaging modality of many, we are interested in sensors better suited to specific tasks. Sensor modality should enter into design of computer vision systems For example, perhaps for flight wide field-of-view sensors are appropriate, and in general useful for mobile robots. [Slide from C. Geyer see Reference Section] 3 Omnivision in Animals • 500 million years ago Trilobites • Diurnal Insects • Nocturnal Insects • Crustaceous • Gigantocypris (not omnivision) 4 Omnivision in Paintings “The wedding of Giovanni Arnolfini” J. Van Eyck 1390-1441 Witch Mirrors 5 Omnivision in Paintings “The praetor and his wife” Q. Metsys 1466-1530 Witch Mirrors 6 Distortion in Paintings Anamorphosis 7 Panoramas (meaning of the XIX century) Patented by Robert Barker - 1787 8 Photography Edinburgh - Courtesy of EdVec Mauna-Kea Observatory 9 10 Wide-Angle Lenses Pros: - Single image [Slide adapted from T. Pajdla see Reference Section] Cons: - Low resolution at periphery 11 Panoramic cameras Cirkut Camera Panning camera Pros: - High resolution per viewing angle Roundshot Swing lens Cons: - slow acquisition; - No dynamic scene - expensive 12 Compound-eye cameras Ladybug PointGrey The Ringcam at Microsoft Research Viewplus Softpia Japan & Gifu University Pros: - High resolution per viewing angle Cons: - Bandwidth - Multiple cameras calibrating and synchronizing; 13 - expensive Compound-eye cameras (inward) Virtualised Reality CMU SMART eMotion (Spin-off Univ. of Padua) now BTS Bioengineering 14 Catadioptric cameras PAL Panoramic Annular lens Catadioprtic Sensor Pros: - Single image Two folded mirror sensor Cons: - Blindspot - Low resolution 15 Catadioptric Camera Composed of: • Standard Camera • Convex Mirror • Support – transparent cylinder – Lateral bar 16 Summary . Omnidirectional sensors come in many varieties ~360º FOV ~180º FOV >180º FOV wide FOV dioptric cameras (e.g. fisheye) polydioptric cameras (e.g. multiple overlapping cameras) catadioptric cameras (e.g. cameras and mirror systems) 17 Omnidirectional cameras Advantages • Wide vision field • One-shot image • High speed • Vertical Lines • Rotational Invariance Disadvantages • Low Resolution • Distortions • Low readability • Customisable field of view • Customisable resolution 18 An omnidirectional camera view Omnidirectional image Panoramic Cylinder (from Centre for Machine Perception, Praha) 19 When is a catadioptric camera equivalent – up to distortion – to a perspective one? h Suppose we are given a catadioptric image For what kinds of mirrors can the image be warped by h into a perspective image? 20 Review: The projection induced by a camera f The projection induced by a camera is the function from space to the image plane, e.g. 21 Review: The projection induced by a camera The projection induced by a camera is the function from space to the image plane, e.g. f-1(p) The least restrictive assumption that can be made about any camera model is that the inverse image of a point is a line in space 22 Review: The projection induced by a camera For many cameras, all such lines do not necessarily intersect in a single point 23 Some optics: Caustics For many cameras, all such lines do not necessarily intersect in a single point Their envelope is called a (dia-)caustic and represents a locus of viewpoints 24 Review: Central projections If all the lines intersect in a single point, then the system has a single effective viewpoint and it is a central projection 25 Review: Central projections If all the lines intersect in a single point, then the system has a single effective viewpoint and it is a central projection If a central projection takes any line in space to a line in the plane, then it must be a perspective projection 26 When is a catadioptric camera equivalent – up to distortion – to a perspective one? If the projection induced by a catadioptric camera is at most a scene independent distortion of a perspective projection, then it must at least be a central projection g 27 When is a catadioptric camera equivalent – up to distortion – to a perspective one? If the projection induced by a catadioptric camera is at most a scene independent distortion of a perspective projection, then it must at least be a central projection The lines in space along which the image is constant intersect in a single effective viewpoint 28 When is a catadioptric camera equivalent – up to distortion – to a perspective one? Question: Which combinations of mirrors and cameras give rise to a system with a single effective viewpoint? 29 Central catadioptric solutions parabolic mirror & orthographic camera hyperbolic mirror & perspective camera elliptic mirror & perspective camera Theorem [Simon Baker & Shree Nayar, CVPR 1998]: A catadioptic camera has a single effective viewpoint if and only if the mirror’s cross-section is a conic section 30 Omni-Vision for Mobile Robots • • • • • Map matching Image-based localization Observation of Optical Flow Biomimetic Behaviours Integration of Omni-vision with other sensors: – Sonar – Laser range finder • • • Outdoor Navigation SLAM (Simultaneous Localization And Mapping) Environment reconstruction & 3D mapping • Miscellanea E. Menegatti - Omni Vision 31 Navigation/Localization Tricks • • • • Invariance of Azimuth Rotational Invariance Vertical Lines mapped in radial lines Robustness to occlusion E. Menegatti - Omni Vision 32 Invariance of Azimuth The azimuth of the object is maintained by the sensor 33 Rotational Invariance Initial Position Image Counter-Rotated P1 Robot Rotated by 90° P2 P5 P4 P3 34 Vertical Lines radial lines Original Image Edge Detection + Hough Transform E. Menegatti - Omni Vision New Design to stretch vertical lines 35 Robustness to occlusion • • Thanks to the wide FOV, usually occluding objects do not change much the image Several similarity measures have been proved to be robust to occlusion • Extreme case presented by Jogan & Leonardis Matjaž Jogan, Aleš Leonardis “Robust localization using an omnidirectional appearance-based subspace model of environment” environment” Robotics and Autonomous Systems 45 (2003) 51–72 36 Applications • • • • • Map matching Image-based localization Observation of Optical Flow Outdoor Navigation SLAM (Simultaneous Localization And Mapping) • Environment reconstruction & 3D mapping 37 Map matching - 1 • Yagi used the vertical edges of the objects to find position of the robot on a map • Edges tracking Y. Yagi, Y. Nishizawa, M. Yachida, Map-Based Navigation for A Mobile Robot with Omnidirectional Image Sensor COPIS, IEEE Trans. Robotics and Automation,pp.634-648,Vol.11,No.5,1995.10 38 Map matching - 2 • Menegatti et al. used the Chromatic Transitions of Interest to perform scan matching • Monte-Carlo Localization Algorithm • Almost the same approach used with Laser range Finders E. Menegatti, A. Pretto, Pretto, A. Scarpa, Scarpa, E. Pagello Omnidirectional vision scan matching for robot localization in dynamic environments IEEE Transactions on Robotics, Vol. Vol. 22, Iss. 3 June 2006 pages 523- 535 39 Image-based navigation - 1 • Ishiguro and Menegatti: – – – – – – FFT magnitude for position FFT phase for heading Self-organization of the memory Image-based Localisation Hierarchical Localization Image-Based Monte Carlo Localisation Emanuele Menegatti, Takashi Maeda, Hiroshi Ishiguro, ``Hierarchycal ``Hierarchycal Image-based Memory for Robot Navigation,'' Navigation,'' Robotics and Autonomous Systems, Elsevier - 2004 Emanuele Menegatti, M. Zoccarato, E. Pagello, H.Ishiguro, ``Image-based ``Image-based Monte-Carlo Localisation with Omnidirectional images'' Robotics and Autonomous Systems, Elsevier - 2004 40 Image-based navigation - 2 • Kröse et al: – Used Principal Component Analysis to extract linear feature – Dataset described in term of eigenimages – Probabilistic localization B. Kröse, N. Vlassis, R. Bunschoten, and Y. Motomura. “A probabilistic model for appareance-based robot localization” localization” Image and Vision Comp, vol. 19(6):pp. 381–391, April 2001.. 41 PCA Explained f " µ + Bc = µ + b1c1 + b2c 2 + ... •f is a vector, e.g., an image •B is a set of basis vectors, the “principal components” •c are the coefficients •Principal Components Analysis discovers the basis vectors one at a time, in the order that minimizes the difference between the reconstructed and the original vectors F, up to rank(F) 42 Image-based navigation - 3 • Gross et al: – Used slices of the panoramic cylinder – Slices confronted via colour histograms – Hybrid map: topological map aumented with metric information T. Wilhelm, H.-J. Böhme, Böhme, and H.-M. Gross. “A multi-modal system for tracking and analyzing faces on a mobile robot” robot” Robotics and Autonomous Systems, 48:31– 48:31–40, August 2004. 43 Observation of Optical Flow • Ishiguro used: – Foci of Expansion (FOE) to estimate relative positions – No encoder info • Svoboda used: – Optical flow to discriminate translation and rotations Tomáš Svoboda, Tomáš Pajdla, and Václav Hlavác. “Motion estimation using central panoramic cameras” IEEE Int. Conf. on Intelligent Vehicles, 1998. Hiroshi Ishiguro, Kenji Ueda and Saburo Tsuji, ``Omnidirectional Visual Information for Navigating a Mobile Robot'', Robot'', IEEE Int. Conf. on Robotics and 44 Automation (ICRA-93), pp. 799-804, 1993. Outdoor Navigation - 1 • Omnidirectional Vision for Road Following with NN: – Road classification – Steering angle Z. Zhu, S. S. Yang, G. G. Xu, Xu, X. X.Lin, Dingji Shi "Fast road classification and orientation estimation using omni-view images and neural networks," IEEE Transaction on Image Processing, Vol. 7, No.8, August 1998, pp. 1182-1197. 45 Outdoor Navigation - 2 Image-based navigation: – – – Topological navigation Histogram matching Different localisation accuracies Paul Blaer and Peter Allen “Topological Mobile Robot Localization Using Fast Vision Techniques” Proceedings of the 2002 IEEE International Conference on Robotics & Automation 2002 46 Outdoor Navigation - 3 Image-based navigation: – Dimension reduction with PCA – Histogram matching José-Joel Gonzalez-Barbosa and Simon Lacroix Rover localization in natural environments by indexing panoramic images Proceedings of the 2002 IEEE International Conference on Robotics & Automation 2002 47 SLAM Michael Kaess and Frank Dellaert, Visual SLAM with a Multi-Camera Rig, Georgia Tech Technical Report GIT-GVU-06-06, 2006 Thomas Lemaire, Simon Lacroix. Long Term SLAM with panoramic vision. Submitted to Journal of Fields Robotics special issue on "SLAM in the Fields". 48 Environment Reconstruction C. Geyer, K. Daniilidis, Mirrors in Motion: Epipolar geometry and motion estimation, Proc. Inter. Conf. on Computer Vision, October, 2003, Nice, France. H.Bakstein H.Bakstein,, T.Pajdla T.Pajdla.. Rendering Novel Views from a Set of Omnidirectional Mosaic Images. Workshop on Omnidirectional Vision and Camera Networks 2003, CD ROM, IEEE June 2003. 49 References WWW: The page of omnidirectional vision http://www.cis.upenn.edu/~kostas/omni.html ICCV03 Course on Omnidirectional Vision http://www.cis.upenn.edu/~kostas/omni/iccv03.html Book: R. Benosman & S.B. Kang (Eds.) Panoramic Vision Springer 2001 50 References Special issues: K. Daniilidis& N. Papanikolopoulos The Big Picture IEEE Robotics & Automation Magazine Dec. 2004 Hiroshi Ishiguro and Ryad Benosman Special issue on omnidirectional vision and its applications Machine Vision and Applications (2003) Vol 14 Yasushi Yagi and Katsushi Ikeuchi Special Issue on Omni-Directional Research in Japan International Journal of Computer Vision Vol. 58, Num. 3, Springer July 2004 Peter Sturm, Tomas Svoboda and Seth Teller Special issue on Omnidirectional Vision and Camera Networks Computer Vision and Image Understanding Vol. 103, Iss. 3, Sept. 2006 51