Department of Computer and Information Science
School of Engineering and Applied Science
University of Pennsylvania
M. Bansal, K. Daniilidis, Joint Spectral Correspondence for Disparate Image Matching, in CVPR 2013 pdf.
N. Atanasov, B. Sankaran, J. Le Ny, Th. Koletschka, G.J. Pappas, and K. Daniilidis, Hypothesis Testing Framework for Active Object Detection, in ICRA 2013, pdf.
Ryan Kennedy, Kostas Daniilidis, Oleg Naroditsky and Camillo J. Taylor, Identifying maximal rigid components in bearing-based localization, in IROS 2012 pdf.
Mayank Bansal, Kostas Daniilidis, and Harpreet Sawhney, Ultra-wide Baseline Facade Matching for Geo-localization, ECCV Workshop on Visual Analysis and Geo-Localization of Large-Scale Imagery, A. Fusiello et al. (Eds.): ECCV 2012 Ws/Demos, Part I, LNCS 7583, pp. 175-186, 2012. pdf.
Butzke, J., Daniilidis, K., Kushleyev, A., Lee, D.D. , Likhachev, M., Phillips, C., Phillips, M., The university of Pennsylvania MAGIC 2010 multi-robot unmanned vehicle system, Journal of Field Robotics, Volume 29, Issue 5, September 2012, Pages 745-761 pdf.
K.G. Derpanis, M. Lecce, K. Daniilidis and R.P. Wildes, Dynamic Scene Understanding: The Role of Orientation Features in Space and Time in Scene Classification, IEEE CVPR 2012.
R. Anati, D. Scaramuzza, K.G. Derpanis, K. Daniilidis, Robot Localization using soft object detection, IEEE ICRA 2012, pdf.
A. Toshev, B. Taskar, K. Daniilidis, Shape-based Object Detection via Boundary Structure Segmentation, Int. J. of Computer Vision, 2012, pdf.
O. Naroditsky, X.S. Zhou, J. Gallier, S.I. Roumeliotis, and K. Daniilidis, Two Efficient Solutions for Visual Odometry Using Directional Correspondence, IEEE Trans. Patt. Anal. Mach. Intell., vol. 34, pp. 812-824, 2012, pdf.
O. Naroditsky and K. Daniilidis, Optimizing Polynomial Solvers for Minimal Geometry Problems, Int Conf Computer Vision, ICCV, 2011. pdf.
C. Phillips, K.G. Derpanis, K. Daniilidis, A Novel Stereoscopic Cue for Figure-Ground Segregation of Semi-Transparent Objects, 1st IEEE Workshop on Challenges and Opportunities in Robot Perception, 2011. pdf.
D. Scaramuzza, A. Censi, K. Daniilidis, Exploiting Motion Priors in Visual Odometry for Vehicle-Mounted Cameras with Non-holonomic Constraints, IROS 2011. pdf.
O. Naroditsky, A. Patterson, K. Daniilidis, Automatic Alignment of a Camera with a Line Scan LIDAR System, In IEEE Int. Conf. Robotics and Automations, 2011, pdf.
A. Makadia and K. Daniilidis. Spherical correlation of visual representations for 3d model retrieval. International Journal of Computer Vision, 2010. pdf.
G.L. Mariottini, F. Morbidi, D. Prattichizzo, N. Vander Val, N. Michael, G.J. Pappas, and K. Daniilidis. Vision-based localization of leader-follower formations. IEEE Trans. Robotics and Automation, 2009 pdf.
N. Moshtagh, N. Michael, A. Jadbabaie, and K. Daniilidis. Vision-based, distributed control laws for motion coordination of nonholonomic robots. IEEE Transactions on Robotics, 2009, pdf.
A. Toshev, A. Makadia, K. Daniilidis: Shape-based Detection of Moving Objects in Videos, IEEE Computer Vision Pattern Recognition CVPR 2009. pdf.
R. Anati, K. Daniilidis. Constructing Topological Maps using Markov Random Fields and Loop-Closure Detection. Neural Information Processing Systems 2009. pdf.
J.P. Tardif, Y. Pavlidis, and K. Daniilidis. Monocular visual odometry in urban environments using an omdirectional camera. In EEE International Conference on Intelligent Robots and Systems, 2008 pdf.
My research field is computer vision and robotics and in particular
space and motion perception with machines. Since 1990, I have
been studying multiple view geometry, image matching,
stereo vision, recognition, sensor deployment, and camera design. Recently, my
group has been working on:
Shape-based Object Detection and Segmentation We address the problem of object detection as a problem of representation and selection. We introduced a new shape-based representation, the chordiogram, and we a new selection procedure that enables simultaneous segmentation and detection by selecting the corresponding image segmets/boundaries. The chordiogram enables the formulation of the problem as quadratic optimization solved with convex relaxation (Toshev et al. CVPR 2010, IJCV 2012).
Image co-salience: We introduced a new matching score that rewards a simultaneous intra-image segmentation and inter-image correspondence between co-salient regions, and optimized it spectrally in a joint graph (CVPR 2007). We applied it in a novel approach for recogition of 3D objects in videos (CVPR 2009).
Minimal problems: We introduced a new optimization of poynomial solvers useful in any RANSAC selection (ICCV 2011) and we found a new minimal solution to structure from motion given directional correspondence (PAMI 2012).
Visual odometry and mapping: Using only omnidirectional cameras and exploiting proximal as well as distal landmarks we have been able to reconstruct one of the longest outdoor maps, a result that attracted immediate attention regarding application in GPS-denied environments (IROS 2008).
Geometric foundations of panoramic cameras and omnidirectional vision: We introduced a novel unifying geometric theory (IJCV 2001) that described reflections of the 3D world on mirrors followed by projections on the omnidirectional plane. This model unifies all second order reflective surfaces and includes the traditional CCD camera as a degenerate case and reveals constraints for self-calibration (PAMI 2002). A representation of lifted points and line projections (circles) in a commo 3D space, called circle space -- after Moebius work; We discovered a group-theoretic characterization of fundamental intrinsic transformations in such systems: they are elements of the Lorentz group O(3,1) (ICCV 2003). This discovery enabled us to estimate motion and intrinsic camera parameters from kernels of 4x4 matrices. Later we generalized the lifting to cover radial distortion models (ICCV 2005).
Image matching: Matching images, for example for location recognition, is challenging when views have small overlap and significant clutter. Using harmonic analysis on groups we have been able to convert a voting scheme for matching without any feature correspondence into a filtering problem, assuming only a consistent scene geometry (PAMI 2003, IJCV 2007, CVPR 2003, 2005, ICCV 2007). We extended our results to range imaging, by finding a procedure for aligning point clouds based on the similarity of their orientation histograms (CVPR 2006).
Vision based formation control: We established the first framework for consensus of a formation of robots using only the lines of sight to neighbors. We proved that this is feasible using only bearing and the time to collision, giving thus a justification to biological evidence that many species flock without having any range sensors (RSS 2005, 2008, Transactions on Robotics 2009). When an agent can play the role of leader, we proved that the formation can act as a collective observer (ICRA 2007).
Tele-immersion and Stereo: In 1998, together with UNC, Brown University, and ANS, we set as our goal to make feasible an immersive sense of remote presence, so compelling that it instantaneously creates the illusion of being near to people and objects which are physically miles away. Our desire to close the loop through networking and graphics made us put a considerable effort on the systems and performance aspect. A series of successful milestones led to a system that acquires an environment as a set of depth maps from several viewpoints in 10Hz (using a dual Intel P4) and can run for several days without interruption (IJCV 2002, IEEE-CSVT 2003).
Program Co-chair ECCV 2010:
11th European Conference on Computer Vision
Watch Sandy Patterson and me for 5min in the Discovery Channel feature
|PhD Students:||Matthieu Lecce , Menglong Zhu , Jason Owens, Cody Phillips , Mayank Bansal , Alexander Patterson, Roy Anati.|
|Alumni PhDs:||Oleg Naroditsky (Ogmento), Alexander Toshev (Google Research), Ankita Kumar, (Oracle), Nima Moshtagh (Lockheed Martin). Ameesh Makadia (Google Research), Volkan Isler (U. of. Minnesota), Christopher Geyer (iRobot), Adnan Ansar (JPL/NASA), Weichuan Yu (HKUST).|
|Alumni Postdocs/Visitors:||Konstantinos Derpanis (Ryerson University), Davide Scaramuzza (University of Zurich), Philippos Mordohai (Stevens Inst. of Technology), Jean-Philippe Tardif (CMU), Irene Cheng (U. of Alberta), Yanis Pavlidis , Rodrigo Carceroni(Google Labs), Rahul Swaminathan (Deutsche Telekom Labs), Xenophon Zampoulis (FORTH/CSI), Joao Barreto (University of Coimbra), Nikhil Kelshikar, Thomas Buelow (Philips Research), Jane Mulligan (University of Colorado at Boulder).|
|Alumni MSE Students:||Barath Sanjaran (USC), Jason Liu (UCLA), Allison Mathis (Lockheed Martin), Dinkar Gupta (2004), href=Daniel Rudoy (2002, Harvard PhD), Andrew Trister (2002, MD-PhD Penn). ,|
Spring 2007, 2008, Fall 2008, 2009, 2010: CIS121 Intro to Programming II
Fall 2006, 2007, Spring 2009, 2010, 2011: CIS580 Machine Perception
Sping 2005: CSE399-002 or cse399b Computer Vision
Fall 2004, Fall 2003, Spring 2002, Fall 2002,: CSE390 Robotics.
Fall 2000, 1999, 1998: CSE240 Introduction to Computer Architecture.
Spring 2004, 2001: CIS 700 Special Topics in Machine Perception
Spring 2003, 2000, 1999: CIS 680 Advanced Topics in Machine Perception
PhD, 1992, University of Karlsruhe with
My advisor's advisor was the
1989 Physics Nobel Laureate
(1913-1993). Wolfgang Paul's advisor was Hans Kopfermann (1895-1963).
Kopfermann's advisor was James Franck (1882-1964)
Nobel Laureate 1925,
Franck's advisor was
Emil Warburg (1846-1931).
Diploma (Masters equivalent) in EE, 1986, National Technical University of Athens
|Address:||Department of Computer and Information Science, University of Pennsylvania, 3330 Walnut Street, Levine Hall 472, Philadelphia, PA 19104. How to reach GRASP|
|Acknowledgments:||Grateful for support through the following grants: NSF-IIP-0742304, NSF-IIP- 0835714, NSF-OIA-1028009, ARL MAST-CTA W911NF-08-2-0004, and ARL RCTA W911NF-10-2-0016, NSF-DGE-0966142.|
Greek name spelling: Κώστας Δανιηλίδης