Panoramic background model for PTZ camera

PTZ cameras have been widely applied in many industrial applications. In the systems of video intelligent surveillance, it is a challenging task to segment and track fore-ground(moving) objects with the PTZ cameras automatically, since the background model cannot be easily computed by the traditional algorithms for fixed camera. In this paper, we present a novel approach for constructing a panoramic background model. Our method consists two stages. (1) To refuse false detection, a clean panoramic background is generated by removing the moving objects. (2) Based on scale invariant feature, current frame can be registered with the background model for foreground subtraction. Then an object-centering tracking system is presented for the surveillance application, in which we demonstrate the advantage of the proposed background model. In the experiments, we quantitatively test our method on many samples.