High-Resolution Multiscale Panoramic Mosaics from Pan-Tilt-Zoom Cameras

We present an automatic software based approach for building extremely high resolution panoramic mosaics from images captured by an off-the-shelf pan tilt zoom camera. Using numerous zoomed-in images from such a camera, existing mosaicing algorithms could theoritically build gigapixel images. However the large number of images that must be processed makes such approaches impractical. Our stratified approach for high resolution panoramic imagery, first constructs a coarse panorama of the scene and then adds in detail from a zooming camera, in a top-down multiresolution fashion. Our approach uses both feature based and direct intensity based image alignment methods. Both the geometric calibration (intrinsic parameters and radial distortion) as well as the photometric calibration and alignment is done automatically. Our fully calibrated panoramas are represented as multi-resolution pyramids of cubemaps. We align hundreds of images captured within a 1-12X zoom range and show results from ...

• Type: Conference
• Year: 2004
• Where: ICVGIP
• Authors: Sudipta N. Sinha, Marc Pollefeys, Seon Joo Kim
• Keywords: ICVGIP 2004, ICVGIP 2008, Images, Resolution Panoramic Mosaics, Tilt Zoom Camera