中文摘要：

现有电子稳像技术主要以平移稳像为主,而对同时存在大角度旋转和抖动的视频稳像能力较差,针对该问题提出一种基于SURF和轨迹滤波的旋转视频稳像算法.首先,采用SURF特征点匹配算法获得相邻2帧匹配的特征点,使用双向最邻近距离比值法剔除误匹配点,利用相似变换模型和RANSAC算法,得到高精度的旋转角度并对当前帧进行反向旋转,从而消除视频旋转.然后,使用KLT特征点跟踪算法生成视频的特征点轨迹.最后,采用Kalman滤波和B样条曲线拟合相结合的方法进行稳像处理,进而得到不旋转且稳定的视频.实验结果表明,该算法适合任何角度旋转的视频,消旋精度达到0.02°,且能够有效地消除视频抖动,对于分辨率低,同时存在运动模糊的视频也有较强的鲁棒性.

英文摘要：

Most existing electronic video stabilization algorithms just work under horizontal and vertical movements and the ability to deal with the video with both large rotation and undesirable shakes is weak; aiming at this problem,a rotational video stabilization method based on SURF （speeded-up robust features） and point-feature trajectory smoothing is proposed.Firstly,the SURF point-feature matching algorithm is employed to find the corresponding matching points between two consecutive frames,then the bidirectional nearest neighbour distance ratio method is used to eliminate false matches; the high accuracy rotation angle of inter-frame is obtained using similarity transformation model and RANSAC （random sample consensus） algorithm,and the video rotation is removed by rotating the current frame reversely.Secondly,the standard KLT （Kanade-Lucas-Tomasi） point-feature tracking algorithm is used to extract the point-feature trajectories from the video.Finally,the Kalman filtering and B-spline curve fitting are combined and adopted to smooth the point-feature trajectories.The stabilized video can be rendered using the new trajectories by full-frame warping.The experiment results confirm that the proposed approach is suitable for the video with any rotation angle,has a rotation removing accuracy of 0.02°,can stabilize the video effectively,and also has a strong robustness for the video with low resolution and motion blur.