快速平滑点特征轨迹电子稳像

Fast smoothing point-feature trajectories for digital image stabilization

为了提高机载成像系统输出视频的图像质量,提出了一种快速平滑点特征轨迹的稳像算法。以消除全局运动估计的帧间匹配累积全局运动、实现长时快速稳像为目的,建立有别于传统实时稳像模式的系统框架。首先采用SURF算法从原始的抖动视频中提取不稳定的特征点;其次利用Delaunay三角剖分算法判断特征点的邻接性,生成点特征轨迹;再次采用Kalman滤波器对不稳视频中得到的点特征轨迹进行滤波处理,得到平滑的点特征轨迹;最后由原始点特征轨迹和平滑点特征轨迹估算出直接需要补偿的全局运动矢量。实验结果表明:该方法不仅能够实时处理失稳航摄视频,有效改善机载成像系统的图像质量,而且能够估计出相互独立的帧间全局运动矢量,可以应用于需要长时间稳像的场合。

To improve the video＇s image quality of airborne imaging systems, we proposed a real-time image stabilization system, based on fast smoothing point-feature trajectories. In this paper, a system framework was established to eliminate the accumulative errors between frames in global motion estimation and achieve fast and long-time video stabilization. This framework was different from the traditional real-time video stabilization. Firstly, an improved SURF algorithm was introduced to extract unstable feature points from original shaky video. Secondly, we determined the adjacency of these feature points, generated point feature trajectories with Delaunay triangulation algorithm and smoothed them with Kalman filtering. Then, we could estimate the global motion vectors which directly needed to be compensated from the original point-feature trajectories and the smoothing point-feature trajectories. The experiment results indicate that the proposed method can be used to stabilize unstable aerial video in real time. Also, it can effectively improve the image quality of the airborne imaging systems and estimate global motion vectors between independent frames. So it can be used in long-time image stabilization.