Fast generation of dynamic and multi-resolution 360° panorama from video sequences

The paper presents a systematic approach to automatic construction of a dynamic and multi-resolution 360° panoramic (DMP) representation from image sequences taken during camera rotation and zoom. Although a simple 2D rigid motion model is used to estimate interframe motion parameters, our mosaicing methodology enables precise cylindrical panoramic mosaic. Moving objects are detected and separated from images based on motion information, and their accurate contours are extracted using a modified active contour algorithm. A multi-resolution representation is built for the more interesting areas by means of camera zooming. The DMP construction method is fast, robust and automatic, achieving 1 Hz on a 266 MHz PC. No camera calibration, feature extraction, image segmentation or complicated nonlinear optimization processing is required in our algorithms. The construction of the panoramic representation can be used in virtual reality and very low bit-rate video coding