Robustní odhad pozice částečně obecné kamery za použití několika minimálních řešičů

Beliansky, Patrik

Camera pose estimation is a fundamental problem in computer vision with various ap- plications, such as in Structure-from-Motion and Visual Localization. These applications often focus on registering a new camera to a set of already registered cameras with known poses. Thus, camera poses can be estimated by solving the semi-generalized relative pose problem. This is usually achieved by using a minimal solver inside a RANSAC-style robust estimator. This estimator can handle data with a large amount of outliers. While solvers for the semi-generalized relative pose problem are typically too inefficient for practical use, the highly related semi-generalized homography problem can be solved effi- ciently. This thesis develops and investigates different implementations, e.g., using Sturm sequences or companion matrix approaches, for multiple solvers for this problem. This thesis further proposes a novel RANSAC-based approach that automatically selects the most suitable solver for each iteration in a data-driven way. In detailed experiments, the solvers are evaluated in terms of efficiency and stability. Extensive experiments show the effectiveness of our novel RANSAC-based approach by comparing it against two baselines. 1

host :: přihlásit Digitální repozitář
		Hledej		Nový záznam		Nápověda		O repozitáři