高精度实时全帧频SURF电子稳像方法

Real-time full-frame digital image stabilization system by SURF

针对视频抖动和电子稳像的实时性,提出了基于SURF的电子稳像算法。首先,采用SURF算法检测图像的兴趣点,建立了当前帧与参考帧的对应关系,得出高精度的运动估计矢量。然后,通过判定参考帧更新策略,获得平滑的帧间全局运动矢量,进而对原始视频序列进行运动补偿。最后,将参考帧对应区域像素填充到稳像帧丢失像素区域进行全帧频补偿,输出高精度的实时全帧频电子稳像视频。实验结果表明,采用SURF算法的实时电子稳像算法,运行时间<30ms,精确度<1 pixel,对存在严重运动模糊的视频具有较强的鲁棒性,可以去除帧间高频抖动并有效改善视频质量。

To overcome the undesirable shakes or jiggles of a camera and to implement the image stabilization in real time,a real-time full-frame video stabilization system based on the Speeded Up Robust Features（SURF） was proposed. Firstly,the SURF was employed to extract feature points ,and the correspondence between the current and reference frame was established to get high precision local motion vector estimation. Secondly, by determining the reference frame update strategy, the smoothed interframe global motion vector was obtained. Finally, the corresponding pixels of the reference frame was filled with a stablized frame to compensate the unstable motion and to output an stablized fullframe video. Experimental results show that the real-time full-frame video stabilization system using SURF algorithm can provide the high accuracy （lower than 1 pixel） and short processing time （less than 30 ms）. Moreover,it has a higher robustness on serious motion-blur and better image quality.