Abstract

This article demonstrates the feasibility of a visual Simultaneous Localization and Mapping (SLAM) algorithm based on the concept of Random Finite Sets (RFS), in which a navigator such as a robot, car or cellphone uses an RGB-D video camera to reconstruct the scene around it and simultaneously estimate its own pose. In contrast to many state-of-the-art SLAM solutions, which rely on fragile map management and measurement-to-map landmark data association methods, the Bayesian based RFS framework circumvents the necessity for such methods. An RFS implementation of Rao-Blackwellized (RB)-Probability Hypothesis Density (PHD)-visual SLAM is presented and its performance is evaluated under various motion, measurement and detection noise levels.