Video Stablelization Based on Grid Wariping and Moving Gaussian Filter

In this article, we propose a method for video stabilization that utilizes grid-based motion extraction and temporal moving Gaussian filtering for smoothing motion. The method determines the parameters for multiple homography estimations using optical flow estimation. Compared to previous methods, it allows for timely adjustment of smoothness and simplifies computational complexity. In experiments, our method is compared with existing 2D and 3D video stabilization techniques. The results indicate that our method approaches the stabilization effectiveness of 2D methods while significantly reducing the computational complexity compared to 3D methods. Furthermore, we find that by fine-tuning parameters such as covariance filtering, homography plane estimation, and the number of motion smoothing iterations, we can further optimize our method to suit various application scenarios.