Research On Dynamic Video And Image Stitching Technology In Large Parallax Scene

Image stitching and video stitching techniques are widely used in panoramic maps,clinical diagnosis,video surveillance,virtual reality and many other fields.Significant progress has been made in image stitching under the single-point perspective assumption,but it is difficult to meet the requirements of the single-point perspective assumption in the image acquisition process,or there are jitters and pose differences between cameras in dynamic video acquisition process,resulting in parallax of interimage is too large,making it difficult to produce clear and natural stitching results.Compared with images,video content is richer and can bring better visual experience to people.Video can be seen as composed of multiple frames of images that are continuous in time.Therefore,video stitching is based on image stitching.For dynamic video stitching,if only the image stitching technique is used to stitch the corresponding frame images of the video,there will be jitter in the stitching result.Therefore,in order to obtain stable video stitching results,the image stabilization problem also needs to be considered.Aiming at the problem that image stitching is prone to blur and ghosting in large parallax scenes,the seam-driven image stitching method is studied,and a large parallax image stitching algorithm based on multi-plane and multi-perceptual stitching is proposed.First,in order to solve the problem of multiple corresponding planes in a large parallax scene,an alignment candidate homography matrix is generated based on multiple planes,and each homography matrix can achieve one-to-one alignment of the planes in the image;at the same time,considering the multi-perceptual characteristics of human eyes on image color,edge and saliency,an image fusion method based on multi-perceptual seam is proposed to generate stitching seam.The seam is used as the stitching boundary to stitch the images together more natural and more seamless.In view of the problem that it is difficult to obtain both stable and seamless stitching results at the same time during dynamic video stitching,a dynamic video stitching algorithm based on semantic segmentation is proposed.The method of simultaneous image stabilization and stitching is used to obtain stable and seamless large-view stitching videos;at the same time,in order to obtain more stable and more accurate video stitching results,a background information extraction algorithm based on semantic segmentation is proposed.The camera motion path is estimated according to the time feature matching points at the background,eliminating the interference of the motion foreground and achieving a smoother video stabilization effect;calculate the image transformation model according to the spatial feature matching points at the background to achieve seamless stitching at the background,and then use the multiperceptual seam fusion algorithm to find a seam in the background area and stitch the image along the seam together,it can also improve the stitching accuracy of video corresponding frame images in the presence of parallax.The thesis has 40 figures,3 tables,71 references.