LARGE-SCALE SCENE VIDEO IMAGE STITCHING METHOD

摘要

Disclosed is a large-scale scene video image stitching method. The method comprises: collecting two pieces of video data whose shooting scenes are overlapped; extracting two background images from the two pieces of video data, calculating a homography transformation matrix between the two background images, and determining an overlapped background area of the two background images by using the homography transformation matrix; and using a background transformation matrix to transform, stitch and fuse the two background images to obtain a background stitching image of a current frame; extracting two foreground images frame by frame from the two pieces of video data; and extracting and performing matching between feature points of the foreground images in the overlapped background area, adjusting image translation parameters in the background transformation matrix according to a position relationship between matching feature points, so as to obtain a transformation matrix of the foreground images, and transforming, stitching and fusing the two foreground images by using a foreground transformation matrix; and synthesizing a foreground stitching image and the background stitching image into a stitching image of the current frame. Embodiments of the present invention can meet the requirement for real-time performance of stitching.