Cooperative hybrid tracking for event monitoring systems

In this thesis, a novel hybrid visual tracking system for event detection and human tracking is proposed. This surveillance system is composed of a stationary camera and a pan tilt zoom (PTZ) camera. The two cameras are geometrically related using camera calibration from images of spheres. By performing a novel sensitivity analysis of the calibration method, guidelines to obtain better calibration results are established. The stationary camera detects events of fall and wandering using motion-based visual tracking. The PTZ camera then tracks and follows the person who triggered an event using color-based particle filtering and PTZ strategies defined in this thesis. The purpose of tracking in view of the PTZ camera is to continuously keep the person in the camera view and to obtain identifying details of the person. Experimental results for camera calibration, event detection, and human tracking are presented to demonstrate the proposed cooperative hybrid visual tracking system.