A Scalable Image-Based Multi-Camera Visual Surveillance System

In this paper, we aim to achieve scalability and wider scene coverage through the use of multiple cameras in an outdoor visual surveillance system. Only image-based information is used to control the cameras, making the system highly scalable. We show that when a Pan-Tilt-Zoom camera pans and tilts, a given image point moves in circular and linear trajectory respectively. We also create a scene model using a directly top-view image of the scene to eliminate perspective foreshortening. The scene model allows us to handle occlusion occurring as a result of line of sight between cameras and tracked objects and to determine the view of an object that a camera is acquiring images of. In addition, we also use a maximum weight matching algorithm to assign cameras to tasks when the number of cameras are limited. Finally, the system is tested by simulations and a prototype system.

• Type: Conference
• Year: 2003
• Where: AVSS
• Authors: Ser-Nam Lim, Larry S. Davis, Ahmed M. Elgammal
• Keywords: AVSS 2003, Cameras, Scene Model, Signal Processing, Wider Scene Coverage