"Event Recognition in Video Streams"

5/16/2003: 10:30am - 12:00pm
11th Floor Large Conference Room

Abstract: It is important to recognize events of interest in dynamic environments for many applications. Video cameras provide a common sensor to view common physical environments. This talk will describe recent work at the computer vision laboratory of USC on automatic recognition of events from video streams. The task is made complicated by the ambiguities inherent in imaging sensors and by the allowed variations, in duration and in style, for the same event. The methods we have deveoped are based on stochastic finite state automata, similar to the hidden Markov model machines used in speech recognition, and based on rigorous Bayesian reasoning. We have also developed an early version of an "event representation language" to simplify the task of specifying desired events for recogntion. This talk will focus on the event recognition methods and touch on the image analysis aspects only lightly. The methods used for higher-level, complex event recognition are relatively independent of sensor modality and should apply to outputs of inferences from a variety of sensors as well.

About Ram Nevatia: R. Nevatia received his Ph.D. from Stanford University. He has been on USC faculty since 1975 and currently is Professor of Computer Science and Electrical Engineering. He is also director of the Institute for Robotics and Intelligent Systems on campus. He teaches courses in computer vision and in artificial intelligence. He has worked on many aspects of computer vision including early processing, object description and recognition and more recently on event recognition.

Last updated: Mon Jun 19 17:44:06 2006