Digital Video Stabilization Algorithm Research

With the advancement of information technology, video recording becomes easyand common in people’s daily life, where video stabilization technologies have beenwidely deployed. Video stabilization technologies can be classified into threecategories, i.e. mechanical, optical, and digital. Among which, as the trends forintegrity and portability, digital stabilization technology has gained extensiveattention for its low cost, high accuracy, as well as its stability. I focus on the studyof digital video stabilization problem in this thesis.Based on computer image process technology, digital video stabilizationalgorithms extract motion parameters of adjacent video frames, and then compensatethem reversely to achieve a stabilized effect. The algorithms usually consist of threeparts, namely inter-frame motion estimation, motion filter and motion compensation.Focusing on the motion estimation problem, which is the central to the stabilizedeffect, I proposed two algorithms in this study.For the video with horizontal movements only, I proposed an algorithm basedon partition selection, after analyzing two types of gray information based algorithm,namely block matching and gray projection. The proposed algorithm divides a videoframe into several subareas first, and quantifies the gray feature of each subarea,then chooses the subarea with strongest feature to deploy the gray projectionalgorithm for motion estimation. Finally implement this algorithm for stabilization experiments using MATLAB.For the video with both horizontal and rotational movements, I analyzed twotypes of local feature based motion estimation algorithm, namely SIFT feature basedand Harris corner matching based. And the Harris corner matching algorithm wasimplemented for rotary video stabilization experiment using MATLAB.