Tracking and handoff between multiple perspective camera views

Abstract

We present a system for tracking objects between multiple uncalibrated widely varying perspective view cameras. The spatial relationships between multiple perspective views are established using a simple setup by using tracks of objects moving in and out of individual camera views. A parameterized Edge of Field of View (EoFOV) map augmented with internal overlap region boundaries is generated based on the detected object trajectories in each view. This EoFOV map is then used to associate multiple objects entering and leaving a particular camera's FOV into and out of another camera view providing uninterrupted object tracking between multiple cameras. The main focus of the paper is robust tracking and handoff of objects between omni-directional and regular narrow FOV surveillance video cameras without the need for formal camera calibration. The system tracks objects in both omni-directional and narrow field camera views employing adaptive background subtraction followed by foreground object segmentation using gradient and region correspondence.