Multi-view Video Stitching Algorithm

With the rise and the continuous improvement of computer technology,high-resolution, high-quality videos gradually penetrate into people's daily life,such as working and learning,as well as video surveillance technology. Since the range of pictrue taken by a single camera is too small and wide-angle pictrue is taken by either the wide-angle canmera which is very expensive, or the fisheye camera with a cost of image distortion, we need to find a technology with a lower price and higher-quality without distortion.And then we got the technology of multi-view video stitching.Multi-view video stitching technology has got a lot of attention in recent years.The multi-view video stitching what this article mentioned is a technology of stitching a set of videos inputs which was taken by videos in different directions and which are already overlapping partly, into a wide-angle panorama with high-quality by using the technology of image registration, image adjusting and so on.This article is a research in video stitching based on multi-veiw cameras.This article has built a framework of muilt-view video stitching with three parts, global registration of multi-veiw cameras, image warp and fusion of multi-veiws cameras, background extraction/objects processing. Video input stream is first separated by background extraction/objects processing to extract background. Then we use these backgrounds to calculate the global registration parameters in the stage of initialization. After several steps, the parameters of global registration will be used in the stage of initialization of image warp and fusion parts. After the initialization, background extraction/objects processing send video input stream directly into the image warp and fusion parts which has been initialized, and doing object precessing on the results which the image warp and fusion outputs.