Active Surveillance Using Dynamic Background Subtraction

A prototype active surveillance system is presented. The system has two cameras, one of which is rigidly xed & has a wide angle lenses while the other is mounted on a pan-tilt head & has a zoomed lenses. The images from the wide angle camera are analyzed using dynamic background subtraction together with normalized cross correlation in order to identify moving people within the scene. An approximate mapping is determined (o -line) between the pan-tilt angles and the position of the zoomed image within the wide-angle view. This mapping is improved (for each frame) through 1-D image based correlation and hence moving people are located in the zoomed view. The system has been tested on a di cult real-world scene and image sequences from both cameras are presented. The potential of the system as an `intelligent' security device and the power of the dynamic background subtration & correlation mechanism are clearly demonstrated.