A Plane-based Indoor Localization Scheme Under Manhattan World Assumption

Localization has been key issue not only in the SLAM(Simultaneous Localization and Mapping)field,but also in applications of LbS(Location-based Service)and AR(Augmented Reality).Among various solutions,visual localization has gained widespread attention and developed rapidly.This paper presents a solution based on plane feature to visual self-localization in indoor environments,using color and range image from RGB-D sensor.In the proposed strategy,indoor scenes are depicted with plane-based map consisting of planar regions.Repetitive subgraph matching process per frame are reduced to one subgraph matching and Manhattan World assumption is used for downscoping in the subgraph matching,which enables more neighbor planes for matching with equivalent time cost.As an evaluation to localization,a plane-based registration process is performed to refine the positioning result and obtain registration score for planes within the subgraph range.The introduction of Manhattan World assumption also helps to simplify the 6DoF(Degree of Freedom)localization after initial positioning to 4DoF localization in subsequent localization.The localization problems in the proposed scheme are grouped into 6DoF positioning in the initial state and 4DoF loop closure problem in the known coordinate.Experiment results show that the proposed scheme requires shorter distance for real-time self-localization than the existing plane-based localization algorithm.The plane-based registration provides localization with the ability of eliminating potential mismatches and holds certain amount of tolerance for dynamic scenery changes,which evidences its robustness to indoor SLAM in practical situations.