Center For Visualization & Virtual Environments
Research Project:
Multi-View Calibration
Investigators:
Christopher Jaynes
Sponsor:
National Science Foundation
Calibration in wide-area video surveillance scenarios is a challenging problem. Outdoor environments are complex, contain stationary and moving objects, and uncontrolled lighting.
We have developed methods that exploit tracked motion trajectories to compute the multi-view camera geometry that relates two (or more) cameras. The approach assumes overlapping fields of view and that cameras are connected via a local network.
A typical scenario for multi-view geometry estimation given a dense cluster of cameras is shown below. Motion is assumed to take place on a locally piecewise planar surface.
