GE Global Research Intelligent Video
Dr. Peter Tu
Abstract
GE Global Research has developed an intelligent video framework targeted at the automatic surveillance
of large sites. The talk begins with a discussion of autocalibration methods enabling the inference of the
location, orientation and intrinsic parameters of a network of cameras with respect to a world coordinate system.
The concept of a One-World-View in which all site imagery is merged into a single view is discussed. A review
of a person detectors and tracking is then presented. Going from gross tracks to articulated motion and gaze
detection is then discussed. Methods for Biometrics at a distance via model fitting and super-resolution of the
face are presented as well as recent work on iris recognition at a distance. Going beyond the location and
identity of an individual, methods for inferring individual intent will be presented. The talk concludes with a
discussion of how all these techniques can be combined into real-time event detection and site-awareness
system of systems.