Efficient Real-Time Localization in Prior Indoor Maps Using Semantic SLAM

Abstract

In this paper, a method for real-time global localization (registration) of an agent (robot or visually impaired person) in the presence of an a priori indoor map (e.g., an image of a floor plan) is presented. In the proposed algorithm, the SLAM map created by the agent equipped with RGB-D and IMU sensors is used in conjunction with an a priori architectural floor plan to find the global location of the agent. This involves extraction and vectorization of the semantic object locations from the a priori map image, implementation of real-time semantic SLAM using onboard sensors on the agent, and the use of a particle filter based optimization method for global localization. The proposed algorithm is applied in an indoor environment and localization results are presented showing the effectiveness of the approach.