Stitching Method For Image Sequence

Studies on image mosaic and fusion have been carried out at home and abroad, and a lot of progress have been made, but most studies are focused on the plane image fusion and splicing. For image sequences with changes in shape, size, and attitude, it is difficult to achieve matching and stitching using the original fully static two-dimensional image with the same size. With depth research on stitching method for image sequence, this paper mainly completed the following work:1. Combined with the existing literature about fusion splicing methods for three-dimensional reconstruction with the image sequences at home and abroad, a summary of the research status and technical difficulties is made.2. Interpreted the key technology of image stitching and fusion. including The processes of image fusion splicing, commonly used feature extraction operator, image registration algorithm, algorithms for matching point purification and transformation matrix estimation, as well as popular image fusion algorithm, are presented. Part of these algorithms are showed through experiments.3. For the shortcomings of traditional image stitching, an image stitching method based on Harris corner were proposed, throuth the understanding of the principle of Harris corner detection, Harris feature matching and weighted image fusion algorithm. Firstly, improved Harris operator with the splines function and circular window search is applied to detect the image corner, with a better localisation performance and effectively avoid the phenomenon of cluster; Then the matching method based on the cross-correlation function, and weight-optimized fusion method after eliminating the false matching; Finally, by comparative experiments of Harris and the proposed method, and analysis of experimental results, the proposed method are verified.4. For the problem of stitching difficult with SIFT algorithm, a stitching method for images of small overlap region was proposed, on the basis of ORB algorithm. Firstly, affine transformation processing are treated on the image, and groups of analog images is formed; then modified ORB algorithm based on pyramid strategy to obtain matching points, combined with improved RANSAC algorithm to eliminate false matching points; Finally, experiments are carried out based on the proposed method, compared with SIFT and the ORB, also analysis at the same time.