PlaneLoc: Probabilistic global localization in 3-D using local planar features.

The global localization problem concerns situations when a map of the environment is known but there is no initial guess of the agent position. Whereas the ability to perform global localization is required in many practical situations, it is still an open problem, particularly if the agent requires to find an accurate estimate of its 3-D pose. In this article, we describe PlaneLoc, a novel probabilistic approach to 3-D global localization, which integrates multiple local cues to construct a probability distribution that describes the likelihood of the agent pose. This framework enables to incorporate various types of localization cues but we demonstrate its feasibility using segmented planes abstracted from RGB-D data. We use multiple triplets of planar segments to generate candidate probability distribution and employ it to find the most probable pose with respect to a global map of planar segments. The PlaneLoc implementation uses the ORB-SLAM2 system that serves as visual odometry and makes it possible to generate observation in a form of sets of local segments online. The proposed approach can be used for global localization with a known map or for loop closing and re-localization in Simultaneous Localization and Mapping. The implemented system is validated in experiments using publicly available RGB-D data sets, including our own data set acquired specifically for testing localization methods based on planar features.