Performance Assessment of Robust & Efficient Video Stabilization Algorithms based on L1- L2 Optimization and s-R-t Transform

Abstract

Video stabilization is the method of removing unwanted movement from a video stream. In this paper, we have proposed three algorithms for stabilization of jittery videos. 1.Video stabilization based on Ll norm 2.Video stabilization based on s-R-t transform 3.Video stabilization based on Ll&L2 norm The first algorithm is based on L1 norm. L-l norm is related with Least Absolute Deviation (LAD). It is minimising sum of absolute difference between consecutive video frames. In the second Algorithm, hybrid technique which is the combination of RANSAC (Random Sample consensus Algorithm) and s-R-t (scale-rotation-translation) transform is proposed to stabilize jittery videos. RANSAC algorithm is used to find effective inlier correspondences and afterward it derives the affine transformation to map the inliers in consecutive video frames. This transformation is capable to improve the image plane. This transform makes smoothening of video frames and also removes jitter in video. To obtain the optimal camera path composed of distinct constant, linear and parabolic segments, we have minimised the first, second, and third derivatives of the resulting camera path. The third algorithm based on LI-L2-norm. L2 optimization achieves the best estimation in least square sense. In order to keep the boundary information of original videos as much as possible optimal smooth camera path should be close to the original path. λ is a weight to adjust the smoothness of path. It can be treated as a factor which controls the degree of stabilization. Comparing the stabilized and shaky video it is confirmed that the processed videos highly satisfy the human perception. Results indicate a remarkable elimination of high jitter from shaky videos.