Comparative Study: Outlier Elimination through Fundamental and Homography Matrices

This paper presents a comparative study between two robust estimation approaches: homography matrix-based RANSAC and fundamental matrix-based RANSAC, for outlier elimination in various computer vision applications. The study focuses on the critical task of reliably estimating correspondences across two-view images. The Random Sample Consensus (RANSAC) algorithm is employed to estimate accurate homography and fundamental matrices robustly, even in the presence of outliers. Image datasets are utilized for experimental analysis, including rotations and translations of object. The performance of both methods is compared in terms of accuracy, robustness based on their geometric properties with the different test dataset. Experimental results demonstrate that the homography matrix-based RANSAC method works well with planar movements of the objects, while the fundamental matrix-based RANSAC method performs better with 3D movements of the objects. The paper concludes by discussing the implications of these findings and highlighting the suitability of each approach.