         
|
GRASP Lab Seminar 2003-2004
April 9, 11:00 AM, Levine Hall 307, hosted by Jianbo Shi.
Larry Davis
University of Maryland
Monitoring Human Activity using Computer Vision
Abstract: During the past decade, we have been studying problems related to detection, tracking and behavioral analysis of humans in action. This talk will review that research, emphasizing our work on real time vision algorithms for visual surveillance. The central tasks of a visual surveillance system are to detect objects (typically people and vehicles) that enter the surveillance site, build models for their appearance so that they can be tracked in both space and time, and model and recognize the interactions of people and vehicles with one another, with fixed objects in the environment and with objects that they transport and exchange. Our research on detection has focused on background modeling and subtraction; I will describe the recently developed codebook algorithm for background modeling, and present a methodology for predicting the performance of background subtraction that is used both as a basis of comparison of our algorithm with others, as well as to identify parameters of the detection process from sample video. We have also developed methods for detecting people from a moving camera platform, and I will describe the components of that research. In the area of continuous tracking we have been studying methods that smoothly combine spatial and color/textural features within a particle filtering framework, and this work will be briefly presented. The persistent tracking problem involves recognizing that an object (person or vehicle) that was viewed previously is the same as an object being viewed now, possibly in another location or under different viewing conditions. I will present a method to represent the color appearance of a person based on color path profiles that have some invariance to pose changes, and explain how they are used to address this persistent tracking problem. Finally, a surveillance system needs methods for specifying the activities that it is to recognize. Such a high level system must cope with uncertain results of image analysis and be able to control the application of vision operators to surveillance video to match models to observations. I will describe an approach to this problem that we are pursuing using Petri nets.
Biography: Larry S. Davis received his B.A. from Colgate University in 1970 and his M. S. and Ph. D. in Computer Science from the University of Maryland in 1974 and 1976 respectively. From 1977-1981 he was an Assistant Professor in the Department of Computer Science at the University of Texas, Austin. He returned to the University of Maryland as an Associate Professor in 1981. From 1985-1994 he was the Director of the University of Maryland Institute for Advanced Computer Studies. He is currently a Professor in the Institute and the Computer Science Department, as well as Chair of the Computer Science Department. He was named a Fellow of the IEEE in 1997. Prof. Davis is known for his research in computer vision and high performance computing. He has published over 100 papers in journals and has supervised over 17 Ph. D. students. He is an Associate Editor of the International Journal of Computer Vision and an area editor for Computer Models for Image Processing: Image Understanding. He has served as program or general chair for most of the field's major conferences and workshops, including the 5th International Conference on Computer Vision, and the 2004 Computer Vision and Pattern Recognition Conference.
