Crowdsourcing Motion Maps based on FootSLAM for Reliable Indoor Pedestrian Navigation in Multistory Environments

FootSLAM is a technique rooted in Bayesian theory that estimates the pose (position and heading) of a pedestrian as he walks in a constrained environment while it simultaneously builds a map of step direction probabilities at the visited areas. To do this, FootSLAM only needs measurements of the pedestrian's steps, a.k.a. human odometry. These measurements may be collected for example by means of a foot-mounted Inertial Measurement Unit (IMU) or using the sensors of a smartphone.

The FeetSLAM algorithm crowdsources FootSLAM by assigning the mapping task to a group of collaborating pedestrians who actively or passively collect step measurements within the same environment. FeetSLAM merges the resulting individual FootSLAM maps to generate a more complete and accurate map of the "walkable" areas.

This thesis extends and improves both FootSLAM and FeetSLAM to bridge the gap towards fully automated crowdsourced indoor mapping. The goal is to provide users with sub-meter accurate maps to improve localization in airports, museums or shopping centers, reduce the time needed to find victims after an emergency call or coordinate a fireghter team during a rescue mission.

The original contributions of this thesis are:
1) extending FootSLAM to multistory environments,
2) reducing FootSLAM's computational complexity from O(t^2) to O(t log t) to facilitate real-time processing of larger areas,
3) reducing the computational complexity of the map combination step of FeetSLAM from quadratic to linear in the area of a floor, and
4) analyzing the requirements and applications of large-scale mapping.