Robust estimation of camera rotation,translation and focal length at high outlier rates

A new approach for robust estimation of camera rotation, translation and focal length at high outlier rates from feature correspondences between two perspective views is introduced. Many computer vision algorithms use RANSAC based methods to achieve robustness. One disadvantage of these methods is the required computational effort at high outlier rates. The proposed approach is based on a new evolutionary global optimization method called Repulsive Particle Swarm Optimization (RPSO). It is shown that the proposed approach requires less computational effort at high outlier rates compared to RANSAC based robust estimation.