3D Object Detection and Pose Estimation for Manipulation: from Single Images to Active Viewpoint Selection
Kostas Daniilidis, Professor of Computer and Information Science, University of Pennsylvania
This seminar addresses the problem of detection and localization of 3-D objects in cluttered scenes. Object exemplars are given in terms of 3-D models without any appearance cues. Deformable part-models are trained on clusters of silhouettes of similar poses and produce hypotheses about possible object locations. Objects are simultaneously segmented and verified inside each hypothesis bounding region using the chordiogram descriptor. A final iteration on the 6-DOF object pose minimizes the distance between the selected image contours and the actual projection of the 3-D model. While research has demonstrated successful grasps based on single images, that selection of class and pose could be further optimized by exploring the capability of active viewpoint selection. When an initial static detection chase identifies an object of interest, several hypotheses are made about its class and orientation, including: planing a sequence of viewpoints, which balances the amount of energy used to move with the chance of identifying the correct hypothesis; formulating an active hypothesis testing problem, which includes camera mobility; and solving it using a point-based approximate POMDP algorithm. Experiments using a 3-D model database and an RGB-D sensor show a significant improvement both in detection and pose estimation. Daniilidis’ research in this area is joint work with Menglong Zhu, Matthieu Lecce, Cody Phillips, Kosta Derpanis, Nikolay Atanasov, Jerome Le Ny, and George Pappas.
Kostas Daniilidis is a professor of Computer and Information Science at the University of Pennsylvania where he has been a faculty member since 1998. He earned his undergraduate degree in electrical engineering from the National Technical University of Athens, 1986, and his Ph.D. in Computer Science from the University of Karlsruhe, 1992, under the supervision of Hans-Hellmut Nagel’s supervision. He was Associate Editor of IEEE Transactions on Pattern Analysis and Machine Intelligence from 2003 to 2007. He founded the series of IEEE Workshops on Omnidirectional Vision. In June 2006, he co-chaired with Pollefeys the Third Symposium on 3D Data Processing, Visualization, and Transmission, and he was program co-chair of the 11th European Conference on Computer Vision in 2010. He has been the director of the interdisciplinary GRASP laboratory from 2008 to 2013 and he has been the Associate Dean for Graduate Education of Penn Engineering since 2013. He is an IEEE Fellow.
This event is part of the Mechanical and Aerospace Seminar Series.