Research On Video Stabilization For Mobile Platform

With the rapid popularity of mobile phones,mirrorless interchangeable lens camera and other portable mobile devices and the popularity of short video applications,recording life anytime and anywhere has become a social trend.Due to the movement of the video shooter and the shaking of the recording device,the final video will have a certain degree of shaking.The main purpose of stabilizing the video,that is,the video stabilization,is to eliminate the jitter in the video,thereby improving the picture rendering effect of the video.In addition,video stabilization can also be used as a preprocessing stage for other video processing,such as object detection,tracking,etc.,to improve the robustness of these tasks.Compared with PC terminals,mobile devices are efficient and portable,and ordinary users are more inclined to perform various tasks on mobile platforms.However,at this stage,there is a lack of technical research on image stabilization suitable for mobile platforms.Compared with PC terminals,the performance of mobile devices lags behind.Therefore,the difficulty of video stabilization on mobile platforms is to ensure the accuracy and immediacy,and to preserve the effective range of the video.This paper focuses on the research on image stabilization technology for mobile platforms.The main work is as follows:（1）Aiming at the high time-consuming problem of video image stabilization,this paper designs a video stabilization model for mobile platforms.Combined with the imaging characteristics of mobile devices and the image transformation model,the video jitter model is simplified into rigid classes and homography classes;The improved optical flow method is used to track video feature points,and the random sampling consistency algorithm is combined to filter the possible interference in the video,so as to ensure that the global motion path of the video can be accurately and efficiently obtained on the mobile platform.Experimental results show that the proposed model effectively reduces the complexity of image stabilization algorithm.（2）A video de-jitter algorithm based on bilateral filtering and norm optimization is proposed.The effects of curve fitting and linear filtering are analyzed.A one-dimensional nonlinear bilateral filtering algorithm is proposed to denoise the video motion path curve with edge preserving.At the same time,a solution algorithm based on l₀ norm optimization is proposed to preserve the video motion path trend,and the two algorithms are combined to denoise the motion path;An adaptive threshold method is designed to preserve the maximum effective region of video;In view of the lack of image stabilization evaluation index in the field of video image stabilization,an image stabilization evaluation index based on gray difference image entropy is proposed.Experimental results show that this method can accurately preserve the subjective motion of video and filter high-frequency jitter.The evaluation index proposed in this paper can make an accurate and objective evaluation of video stabilization effect.（3）Aiming at the common problems of edge clipping and black edge in video image stabilization processing,a video edge restoration method based on panoramic image mosaic is proposed.According to the panoramic image shooting process,the video frame transformation is equivalent to the panoramic stitching process,and the adjacent video frames are stitched by the method based on feature point matching;According to the characteristics of video jitter,a method of segmented splicing of video is proposed.Experimental results show that this method can effectively solve the problem of edge missing caused by video image stabilization.