Video Stabilization Algorithm Design Based On Panoramic Stitching

A single camera only provides limited information on account of the narrow viewing angle.In daily life,considering the needs of security,convenience and entertainment,it's necessary to obtain multiple camera perspectives.Image stitching is to splice images from different views of the same scene,and blends the overlap of different perspective images,so as to expand the viewing angle.Panorama video stitching is the expansion of image mosaic.It stitches multiple perspective video frames of the same time and the same scene to obtain the panorama video.Panorama video has a wide range of applications,including tourist attraction,commercial promotion,panoramic monitoring,panorama live,medical research and so on.Panorama mosaic,firstly calibrates the camera to get the camera's internal parameters,external parameters and distortion coefficient,and uses distortion factor to do the correction of image distortion.Then it calculates the relative position between the cameras.Video frames rotate and shift according to the relative position,so that the same region of the image overlaps and the angle of view expands.Due to the problems of exposure difference and incomplete alignment,the quality of mosaic is not good enough.So it's necessary to do multi-camera exposure adjustment and image blending.The exposure adjustment is to make the exposure of different angle images consistent and ensure the color consistency of the panorama.The image blending of overlaps is to further eliminate the stitching marks,which makes the transition between stitching images more natural.The video captured by hand-held equipment is often trembling,which causes the viewer's motion sickness,and it is more obvious in panorama video.So we need to eliminate the video jitter.Panoramic video is one kind of distorted video,so it's not suitable to use undistorted video stabilization algorithm.In this thesis,we propose a more general stabilization algorithm,which is effective for both distorted and undistorted video stabilization.In order to ensure the quality of video stabilization,we need a sufficient number of tracking features,and the feature distribution should also be more uniform.In this thesis,by using a grid-based optical flow tracking method,the threshold value of feature detection of each grid is adjusted according to the texture information so that the region with less texture can also get enough feature,which guarantees the denseness and uniformity of the feature.In addition to the feature point tracking,we also join the line tracking to improve the robustness of the algorithm,which not only can make up for the problem that the scene can only track a small number points,but also can play a role in maintaining the shape of the object.There are already abundant point tracking and line tracking information,and finally the video stabilization is done based on the mesh model.Through a large number of experiments,our algorithm can be applied to both the undistorted and distorted video for stabilization.