Context-Aware Visual Search Using a Pan-Tilt-Zoom Camera

Author(s):

Philip David, André Harrison, Ramavarapu Sreenivas, and Joshua Whitman

Abstract:

There is a need to perform visual surveillance in large indoor and outdoor environments where regions of interest (ROIs) may span horizontal views up to 360◦ and possess very large depths of field (e.g., 1 km or more). While in some scenarios an Internet-of-Things (IoT) approach allows multiple distributed cameras to cooperatively survey a large ROI, other scenarios, such as surveillance by a mobile robot, require the ROI to be surveyed by a small number of collocated cameras. In the latter scenarios, existing dual-camera methods that use a wide-angle camera to cue a pan-tilt-zoom (PTZ) camera are inadequate as the system is unable to observe very distant objects and events of interest in the wide-angle camera imagery, and therefore unable to cue the PTZ camera. To our knowledge, we are the first to address the surveillance problem of resource-constrained PTZ camera search for objects unobservable in wide field of view imagery. In this report, we first motivate and formally define the problem. We next propose solutions that use a single PTZ camera to intelligently and efficiently search a large ROI for objects of interest. Finally, we present results of experiments performed in simulated 3D environments.