Robust Pan, Tilt and Zoom Estimation for PTZ Camera by Using Meta Data and/or Frame-to-Frame Correspondences

Abstract

An algorithm to estimate pan, tilt and zoom (PTZ) parameters of a PTZ camera from meta data and frame-to-frame (F2F) correspondences at different sampling rates is proposed in a real-time video surveillance and automatic object tracking system. Two extended Kalman filters are designed to simultaneously estimate zoom and pan-tilt parameters. Uncorrelated constant velocity models are used to model the kinematics of focal length, pan and tilt motions, while the F2F homography is employed to model the relative motion of the camera. Experiment results from both synthetic and realtime system data demonstrate that the F2F correspondence information can enhance the PTZ estimation accuracy as long as its error is smaller than a particular threshold