Precise 3D foot-mounted indoor localization system using commercial sensors and map matching approach

Acquiring precise and reliable 3D pedestrian trajectories is gradually developed into an essential task for achieving indoor location-based services. A foot-mounted positioning system (FPS) is proved to be an effective approach for multi-floor indoor navigation, while the performance of FPS is usually reduced by the cumulative sensor error, disturbed local magnetic field, and external accelerations. This paper proposes a precise 3D foot-mounted indoor localization system based on the commercial sensors and map matching approach (3D-FSMM). The 1D convolutional neural network model is applied to the detection of quasi-static period to enhance accuracy of the zero velocity update technology algorithm, and the multi-level observations are extracted to constrain the positioning error originated from the low-cost inertial sensors and complex local environments. In addition, the indoor map information is further extracted for corner detection and optimization of estimated trajectory, and the error ellipse is established for indoor map matching in order to provide more absolute reference. The experimental results indicate that the proposed 3D-FSMM realizes meter-level positioning accuracy in disturbed and multi-floor contained indoor scenes, and has the potential for large-scaled indoor applications.