A framework of surveillance system using a PTZ camera

Abstract

For enlarging the surveillance area, more and more visual surveillance systems exploit Pan Tilt Zoom (PTZ) camera. This paper proposes a framework of surveillance system which uses a single PTZ camera. The framework is divided into two phases: offline phase and inline phase. During the offline phase, camera parameters for every image are computed using SIFT features and bundle adjustment algorithm, and then the mosaic and the background model of the whole area are generated based on the camera parameters. During the inline phase, the real-time frame is projected to the correct location on the mosaic using SIFT features and bundle adjustment algorithm, and then the moving object is detected by background subtraction technical. Experiments show that the PTZ camera's parameters can be computed in time and the moving object can be detected perfectly even when the zoom value changes a lot.