Indoor Localization through Pedestrian Dead Reckoning and Activity Recognition

Indoor Localization through Pedestrian Dead Reckoning and Activity Recognition: A smartphone-based proof of concept

van Ingen, Bart (TU Delft Mechanical, Maritime and Materials Engineering)

Kok, M. (mentor)

Delft University of Technology

Mechanical Engineering | Systems and Control

2020-12-03

Pedestrian indoor localization is a problem yet to have a solution that is as generally accepted as the Global Positioning System is for outdoor localization. An infrastructure-independent option is to leverage the data from Inertial Measurement Unit (IMU) sensors to determine position. An advantage of this approach is that these sensors are already available to many people as they are often integrated into the modern smartphone, a ubiquitous device in the developed world. Creating an indoor localization solution that works on this device could, therefore, lead to widespread adoption. This has been recognized by the research community, with methods already being tested on the device. An interesting combination worth considering is that of smartphone-based indoor localization with the emerging market of wearable technology. This new technology has the potential to provide additional information that could improve indoor localization. One option is to detect activity that is not easily
detectable from smartphones alone. Within this thesis, a proof of concept is designed with the goal of finding a way in which activity recognition from wearable tech can help smartphone-based indoor localization. The implementation within this thesis relies only on the IMU data of a smartwatch and smartphone, and map information of the indoor environment.
The system consists of two subsystems placed in series, where the output of the first subsystem, the Step and Heading System (SHS), is the input of the second subsystem, the Particle Filter with spatial context. The SHS uses IMU data from the smartphone to determine if a step is taken, what its length is, and in what direction it was taken. The Particle Filter with spatial context uses the resulting SHS trajectory to generate a position estimate in an indoor environment. It uses spatial constraints, such as walls, to limit the estimate. The IMU data of a smartwatch is used to detect interaction with doors. These interactions in combination with known door locations allows the particle filter to calibrate the position estimates to door locations. In order to evaluate performance, experiments were performed in an indoor environment of which map information was available. During the experiment six different paths were walked, while interacting with doors in the environment. Smartphone and smartwatch IMU data were recorded. Door interactions were also recorded manually for reference. Results from the experiments show that, although false positives in detection can have a detrimental effect on
the position estimate, activity recognition can improve position estimates for a smartphone-based system using the implemented design.

indoor localization
particle filter
step and heading system
smartphone
IMU sensor fusion

http://resolver.tudelft.nl/uuid:f41464e8-b192-483b-a41a-9f947b876243

Student theses

master thesis

© 2020 Bart van Ingen