Self-Contained Indoor Pedestrian Navigation Using Smartphone Sensors and Magnetic Features

Abstract

This paper presents an algorithm for navigating in challenging indoor environments that do not have WiFi. Dead-reckoning (DR) based on off-the-shelf smartphone sensors and magnetic matching (MM) based on indoor magnetic features are integrated. For DR, we utilize a two-filter algorithm structure and multi-level constraints to navigate under different human motion conditions. For MM, we use several approaches to enhance its performance. These approaches include multi-dimensional dynamic time warping, weighted $k$ -nearest neighbor, and utilization of magnetic gradient fingerprints. Furthermore, realized that the key to enhance the DR/MM performance is to mitigate the impact of MM mismatches, we introduced and evaluated two mismatch-detection approaches, including a threshold-based method that sets the measurement noises of MM positions based on their distances to the historical DR/MM position solutions, and an adaptive Kalman filter-based method that introduces the estimation of the innovation sequence covariance into the calculation of the gain matrix instead of adjusting the measurement noises. The proposed mismatch-detection mechanism reduced the DR/MM errors by 45.9%–67.9% in indoor tests with two smartphones, in two buildings, and under four motion conditions.