Activity Semantics-Based Indoor Localization Using Smartphones

Abstract

This article presents an activity semantics-based indoor localization approach using smartphones. The activities of pedestrians consist of several continuous activities during the walking process, such as turning at a corner. In our approach, we first use deep learning (DL)-based pedestrian dead reckoning (PDR) to obtain the velocities and distances of pedestrians adopting multiple usage modes (texting, swinging, bag, and pocket), detect pedestrians’ activities in these multiple usage modes, and finally obtain the activities and displacements of pedestrians. Second, we build a topological map composed of all activities in the planar indoor route network and perform shape matching between the topological map and the displacements in the detected activity sequences, gradually converging to the real trajectories of the pedestrians in the route network. Finally, the real trajectories and displacements obtained in this data-driven manner are fused via factor graph optimization (FGO). With the proposed method, pedestrians moving in indoor environments with unknown initial points can be precisely localized. The proposed method is evaluated in an office building, and the results demonstrate that the proposed method can achieve automatic pedestrian localization when the initial points are unknown, with a higher average accuracy than the traditional PDR methods, and can adapt to the localization of pedestrians adopting multiple smartphone usage modes.