Detection and compression of moving objects based on new panoramic image modeling

In this paper, a parametric video coding method based on new panoramic modeling is proposed for panning cameras. An input video frame from a panning camera is decomposed into a background image, rectangular moving object regions, and a residual image. Each area is then coded separately. In coding the background, we employ a panoramic model that can account for several image formation processes, such as perspective projection, lens distortion, vignetting and illumination effects. Moving objects are detected, and their minimum bounding rectangular regions are coded using a JPEG-2000 coder. The reconstruction error using only the estimated background and the moving objects is computed, and the residual image is separately encoded for image quality enhancement and rate control. We evaluated the effectiveness of the proposed algorithm using several indoor and outdoor sequences and found that the peak signal-to-noise ratio (PSNR) improved by 1.3∼4.4 dB compared to that of JPEG-2000.