OS‐ELM‐based visual and inertial indoor localisation method with adaptive matching step

Abstract

Indoor localisation has been a hot research topic because of its importance for location-based services. The proposed method takes affine transformation matrices and inertial information as input vectors to online sequential extreme learning machine (OS-ELM) and outputs relative translation between the two matching frames. The universal approximation capability and extreme learning speed of OS-ELM enable the proposed method to adapt to several matching steps and reduce the localisation error to a low scale. Fuzzy multiobjective decision making and corner detection are also used to improve the final result. Experimental results validate that the proposed method can provide a good localisation accuracy and improve the efficiency of localisation.