Research On Structure Preserving Based Video Stabilization Method

With the development of science and technology,electronic devices are becoming more and more popular in our daily life,so we can capture videos easily by variety of electronic devices.However,electronic device may cause image jitter or image in-stability due to the mechanical vibration of the carrier,therefore,video stabilization techniques come into being.The purpose of video stabilization is to remove the jittery part of video and obtain visually stable result.This paper focuses on the theme of video stabilization.The main content of this paper is as follows:(1)This paper introduces the classification of current video stabilization methods,the current status of application of video stabilization at home and abroad,research sta-tus of electronic image stabilization technology and the theory of 2D video stabilization method.(2)This paper presents a 2D video stabilization method based on multi-path op-timization.First,we estimate motion between two adjacent frames.Each frame of the input video is divided into an uniform mesh,and the motion of each of mesh is estimated.Then,we smooth the camera path.The camera path is calculated by accu-mulating the motions estimated the previous step,and an multi-path optimization based path smoothing model is designed to smooth the camera path.Finally,we warp each frame of the input video.The comparison shows that the proposed method is efficient and can produce satisfactory results.(3)This paper presents a novel video stabilization method such that the structures of objects in the output video are well preserved.To achieve this target,camera mo-tion is estimated between two adjacent frames,and presented by MeshFlow,which is defined as the motion vectors at the mesh vertices.To efficiently estimate reliable mo-tion vectors,the feature points are clustered into several classes,and the motion vec-tor is estimated for each class.The motion vectors are then propagated by the "as-similar-as-possible" regularization and spatial smoothness term.where the "as-similar-as-possible" regularization preserves the structures,and the spatial smoothness term makes the motion vectors be smooth.In the path smoothing step,the structure preserv-ing regularization term is also applied to reduce the geometric distortion.We compare our method with state-of-the-art methods and two well-known commercial softwares,the results demonstrate that the computation speed of our method is much faster than other methods,and the results produced by our method have a higher cropping rate and distortion degree,which means less content loss and less distortion.