Abstract: To solve the problem that now there is no method that can remove high frequency noise and low frequency noise simultaneously, this paper proposed a new method combining filtering and curve fitting, which can solve the high frequency noise and low frequency noise problem simultaneously. Firstly the offset between frames was estimated fast using bit-plane matching algorithm in this new method. Secondly the inter-frame offset was cumulated to get the global motion vector relative to the reference frame. And the global motion vector was filtered with Kalman filter to remove the high frequency noise. Finally curve fitting was executed on the global motion vector that has been filtered with Kalman filter to remove the low frequency noise. At last the steady intentional trajectory was got. The experimental results show that the method can effectively remove the high frequency and low frequency noise simultaneously, and video stabilization effect is good.

Key words: Kalman filtering, Curve fitting, Diamond search, Initiative motion estimation, Motion Compensation

摘要： 针对目前电子稳像算法无法同时去除高频噪声与低频噪声的问题,提出了可同时去除高频噪声和低频噪声的滤波与曲线拟合相结合的方法。该方法首先用位平面匹配算法快速估计出帧间的偏移量;其次对帧间偏移量进行累加,计算出当前帧相对于参考帧的全局运动量,并对全局运动量进行卡尔曼滤波,以去除高频噪声;最后,对卡尔曼滤波的结果进行曲线拟合以去除低频噪声;最终,得到稳定的主运动轨迹。实验证明,该方法可以有效地去除高频和低频噪声,视频稳定效果良好。

关键词: 卡尔曼滤波, 曲线拟合, 钻石搜索, 主运动估计, 运动补偿