中文摘要：

提出一种基于优化梯度滤波的全局航拍视频配准算法。该方法首先提出一种基于优化梯度滤波的Hession检测器,以实现特征点的精确定位,同时,为了保证在不同摄像机焦距下获得相同的特征点,保留3个分辨率下具有恒定不变的特征点。然后利用最小生成树方法对待配准点进行初始匹配。一致特征点建立后,通过利用非线性最小二乘（NLLS）和随机采样一致性（RANSAC）算法选取具有全局最小误差的变换参数对视频帧间实现配准。实验结果表明,通过利用优化梯度滤波和全局最优模型估计可实现帧间的精确配准,对不同动态场景和光照变换具有较强的适应性。

英文摘要：

A global aerial-video registration algorithm using the optimal derivative filters is proposed.Firstly,the Hession detector based on optimal derivative filters is presented to determine the location of feature points.To ensure that the same feature points are detected in images with different focus from camera,we choose feature points that constant across the three resolutions.Then the minimum spanning tree（MST） is used to find initial matching.Once matching feature points have been found,the transformation parameters which have global minimum error are then estimated using non-linear least squares（NLLS）and random sample consensus（RANSAC）method.Finally the image registration is finished.Experimental results show that by accurately registering frames at the background using optimal derivative filter and globally-optimal transformation model,the technique is robust under different dynamic scenes and illuminations.