Sequence-based Magnetic Loop Closures for Automated Signal Surveying

Ubiquitous indoor positioning remains a challenge for academia and industry alike. Techniques based on signal surveys (fingerprint maps) are arguably the most prevalent today, but these rely on time-consuming, complex manual surveys of the signal propagation. Recently progress has been made in semi-automated surveying, where Pedestrian Dead Reckoning (PDR) algorithms are combined with signal measurements using Simultaneous Localisation and Mapping (SLAM) techniques first developed in robotics. In this work we advocate the use of a dedicated surveyor whose trajectory is estimated using a SLAM based PDR corrected using magnetic signals. Magnetic signals are fast varying indoors, but have poor spatial locality. We show how to match sequences of magnetic measurements within a survey walk in order to identify loops in the trajectory for the SLAM algorithm. Having estimated a trajectory, we build WiFi and Bluetooth Low Energy (BLE) signal maps using Gaussian Processes regression. Magnetic maps are not used for positioning. We test our system in a range of testbeds. We find magnetic SLAM to estimate the trajectory of a 10 minute walk to high accuracy (average error of any point 73 cm, subjective-objective error 22 cm). The resultant signal maps are found to give good positioning accuracies within 3-4 m of the trajectory and come close to the manual survey accuracies, despite requiring significantly less manual effort. We believe our system offers a robust and easy way to survey an environment.