运动背景下高精度视频稳定算法仿真研究

Research on Simulation of Highly Accurate Video Stabilization Algorithm under Moving Background

针对传统视频稳定算法存在的前景局部运动影响全局运动估计精度的难点问题,提出了一种可靠特征集合匹配的高精度视频稳定算法。算法首先利用距离筛选法和金子塔模型提高特征匹配精度,然后利用MLESAC算法剔除运动物体上的无效特征,最后将保留下来的精确匹配特征带入仿射运动模型求出全局运动矢量,并据此对视频帧进行运动补偿以实现稳定视频的目的。实验结果表明,相比于传统基于特征匹配的视频稳定算法,改进全局运动估计算法提高了视频稳定的精度。

Aiming at the the problems of traditional video stabilization that the global motion estimation accuracy is reduced by the local motion. A highly accurate video stabilization algorithm based on feature matching was put forward. Firstly screening method based on distance and Gaussian pyramid model was used to improve the feature matching accuracy, and then MLESAC algorithm was used to remove the features from local motion object. Finally, the global motion estimation vector based on affine model was computed by accurately matching preserved features and the trembling image was compensated to stabilize the image sequence according to the global motion estimation vector. The experiment results show that compared with the traditional feature-based matching video stabilization algorithm, the motion estimation algorithm improves the video stabilization accuracy.