On Sampling Focal Length Values to Solve the Absolute Pose Problem

摘要: Estimating the absolute pose of a camera relative to 3D representation scene is fundamental step in many geometric Computer Vision applications. When calibrated, can be computed very efficiently. If calibration unknown, problem becomes much harder, resulting slower solvers or requiring more samples and thus significantly longer run-times for RANSAC. In this paper, we challenge notion that using minimal always optimal propose compute with unknown focal length by randomly sampling value an efficient solver now calibrated camera. Our main contribution novel scheme enables us guide process towards promising values avoids considering all possible once good found. The RANSAC variant faster than current state-of-the-art solvers, especially low inlier ratios, while achieving similar better accuracy.