基于全局仿射变换估计的航拍视频校正

Global affine motion estimation for aerial video registration

提出了一种快速的基于特征点匹配的全局仿射运动估计方法,用于航拍视频校正和运动检测。为建立对应点集合,改进的Harris角点检测器用于从图像序列中提取和选择稳定的角点,并采用SURF描述子描述这些角点。在运用随机一致性采样方法求取运动参数之前,将匹配对视为矢量进行分析,滤除明显的误匹配对以提高内点率。结果表明,该方法可实时、准确地估计全局仿射运动,完全能够满足移动平台下运动检测的需要。

This paper proposed a fast feature-based global affine motion estimation method for aerial video registration and motion detection.To establish the correspondences,developed the improved Harris corner detector to extract and select stable corners from image sequences,and extracted the SURF descriptor to describe those corners.Ahead of utilizing the random sample consensus technique for motion parameters estimation,viewed the matched couples as vectors,and analyzed to filter out the obvious mismatches to improve the ratio of inliers.Experiments demonstrate that the presented method can robustly estimate the global affine motion parameters with high accuracy,and can meet the requirements of motion detection for mobile platform.