The Video Stabilization Algorithm Based On MEMS Gyroscope

The rapid development of portable electronic devices has enriched the collection forms of image information,such as smart phones,smart unmanned aerial vehicles,and driving recorders.Although the image information collection is convenient,due to the weight of the portable electronic devices is lighter,and the stability is poorer,vulnerable to the interference of noise,causing problems such as video will be jitter and distortion.Therefore,the video must be stabilized to better exert the application value of the electronic video stabilization technology.With the research on the electronic video stabilization technology,from the early 2D stabilization to the 3D stabilization,the stabilization technology has made great progress,but there are still some problems that need to be improved.In this thesis,the research is conducted from the aspects of the rotation quantity estimation of camera platform,image pixel translation quantity estimation,image motion compensation and optimization,aiming to avoid problems such as matching errors of feature matching methods and the complex calculation of 3D motion scene recovery methods.The main work of the thesis is summarized as follows:1.A method which estimating and compensating the rotation jitter based on MEMS(Micro-Electro-Mechanical System)gyroscope is proposed.In order to solve the complicated problem caused by the use of the 3D motion model in the existing 3D video stabilization methods.The MEMS gyroscope is used to estimate motion vectors.The method is proposed based on the parameters calibration of gyroscope,camera calibration and data filtering of gyroscope,in order to gain more precise camera platform rotation motion vectors.In addition,the method can avoid the matching error caused by the single background and fewer feature points,and at the same time reduce the calculation of 3D methods.Experiments show that the efficiency of the algorithm is improved more than 55 percent compared to 3D methods.2.A method which estimating and compensating the pan jitter based on the combination of Harris corner detection and block matching is proposed.Because camera depth can lead to the parallax problem and gyro-solutions cannot get accurate pan jitter information of the camera platform,using the image vision method to analyze the unstable image.The method is proposed based on the Harris detector and the image block matching,calculating pixel translation by the transform model among the adjacent images,and avoiding the recovery of camera platform translational jitter information directly.Experiments show that the accuracy of the algorithm is improved more than 13 percent compared to the gyro-solutions and 3D methods.3.A method which filling the edge black region based on image mosaic is proposed.For the common method of image cutting causes image edge information loss,lower image resolution,as well as there is obvious splicing trace in the image mosaic method.In order to solve the problem,the registration of image sequences is done by using the mapping relationship between the adjacent images,and image mosaic is completed.The proposed method can dynamically blended the overlapping area of the image through the weight function,so that the pixels at the image mosaic edge can be transferred as smoothly as possible,and achieving a better mosaic effect.