Context-Aware Pan-Tilt-Zoom Cameras
This project explores the possibilities of combining Pan-Tilt-Zoom (PTZ) cameras omni-directional sensors and sensor networks. The goal is to create a fully autonomous PTZ system capable of reacting to the context to generate informative video streams with little or no human intervention. The project will explore PTZ control schemes, context sensing modalities, and also associated visualization technology.
Background & Objective: Pan-Tilt-Zoom (PTZ) camera system provides the ability to recover unparalleled visual detail. However, current PTZ systems have very a limited field of view, and therefore depend on significant attention from a human operator to supply the contextual information necessary to realize the full potential of the sensor platform. This reliance on timely and skilled human intervention significantly limits the potential deployment of PTZ camera hardware. The goal of this project is to create a drop-in replacement for legacy, fixed camera systems that provide the advantages of upgrading to a PTZ system without the prohibitive cost of wiring and staffing that is normally associated with a PTZ installation.
Reviewing recorded video from a PTZ camera is a difficult task, particularly if the camera is not under human control. This project therefore also needs to address the problem of presenting PTZ video in an intelligible format. This presentation format should allow the operator to understand both the context that the PTZ is operating in and well as the specific information the PTZ is providing about the scene.
Technical Discussion: So far we have investigated the problem of automatic calibration. This is essential to reducing installation costs. We have built research prototypes that automatically calibrate PTZ camera to omnidirectional cameras and to sensor networks.
The problem of calibrating PTZ cameras to Sensor networks has forced us to consider a new approach: functional calibration. In functional calibration we seek to recover a description of the relationship between the camera and the sensor nodes that will allow us to make the best use of the PTZ camera. The system directly learns a policy that allows the PTZ camera to capture high-quality video of targets that are sensed by the network. This approach allows us to mix sensors with very different modalities: cameras and motion detectors, for example.
Publications:
Technology Area: Computer Vision
Modification Date: November 18, 2008