Joint Colloquium Distinguished Lecture Series
Visual Grounding for Robots and Mobile Interfaces
Wednesday, September 24, 2008
306 Soda Hall (HP Auditorium)
4:00 - 5:00 pm
Department of Electrical Engineering and Computer Sciences, UC Berkeley
and the International Computer Science Institute
Interaction with computer systems is becoming an increasingly mobile affair, and therefore understanding the physical context of queries or commands is an important challenge. Recent advances in computer vision make this possible with relatively low-cost cameras and computation resources. In this talk I'll present methods my group has recently developed towards this goal; I'll describe new techniques for human motion tracking, gesture understanding in dialog, and visual object recognition. In particular we have developed a new probabilistic regression scheme for modeling the image-to-pose mapping that can be efficiently used with very large training sets, a technique for learning shared visual word representations in multi-task visual category recognition, and a multimodal method for learning visual models of particular word senses as defined in various online dictionaries. Throughout the talk I will show example applications involving human-robot interaction and perceptually grounded information search from mobile devices such as cell phones and PDAs.
Prof. Darrell recently moved to UC Berkeley from MIT, where he was a member of the EECS department and CSAIL laboratory. Prior to that he was on the research staff at Interval Research Corporation in Palo Alto, CA. He received the Ph.D. and S.M. degrees from MIT and a B.S.E. from the University of Pennsylvania. He serves on the Editorial Board of IEEE PAMI and the Artificial Intelligence Journal, recently chaired a DARPA ISAT study on video exploitation, and is co-Program Chair for CVPR'10. His interests include computer vision, machine learning, and multimodal human-computer interaction.
Lecture: streaming vieo
|Return to EECS Joint Colloquium|