Algorithm for Stitching of Video Sequences

B. Keck, S.S.G. Rakash, and U. Zoelzer (Germany)


Image planes, matching points, 3D planes


A combination of two or more adjacent images - stitched together - is called a panoramic image. Nowadays panoramic images are becoming popular due to its in creased field of view (often 360). In this paper, we propose a method for stitching two video sequences based on image stitching. The important steps in image stitching are the im age matching and image transformation. We use Scale In variant Feature Transform (SIFT) for image matching. The image transformation is done based on the difference in the image planes. For stitching two videos, these two steps can be repeated for each frame. But this consumes more time. Instead, we reuse the stitching information of the previous frames to speed up the process. The re-usage mechanism is valid only as long as the two video cameras have no rel ative movements. A method has been developed to detect the relative movement between the cameras and the recal culation is done automatically. Secondly, we propose an algorithm for constructing the rectilinear panoramic image of long objects(like train), from video of such objects.

Important Links:

Go Back