RANSAC Algorithm for Matching Inlier Correspondences in Video Stabilization

Abstract

In order to stabilize a video sequence we need to find a transformation which reduces the distortion between frames. To find this transformation feature points must be identified in consecutive frames. In order to get the correspondences between feature points Sum of Squared Differences (SSD) is adopted as matching cost between respective points but by this technique, many of the point correspondences are obtained and they have limited accuracy. To rectify this dilemma, Random Sample Consensus (RANSAC) algorithm is used which is implemented in the Geometric Transform function in Matlab. Utilizing the Random Sample Consensus (RANSAC) algorithm, a robust estimate of transformation between consecutive video frames could possibly be derived. In this paper RANSAC algorithm can be 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 .The RANSAC algorithm is repeated multiple times and at each run the cost of the end result is calculated via Sum of Absolute Differences between both image frames. SAD measures the distortion between two frames by evaluating the similarity between image blocks. On the cornerstone of SAD values, affine transform is obtained which makes the inliers from the initial set of points to match with the inliers from the following set. It is clear from simulation results, inliers correspondences gets exactly coincident which gives more favorable results. The cores of the images are generally well aligned. Thus by utilizing the Random Sample Consensus (RANSAC) algorithm, a robust estimate of transformation is obtained.