Visual-Inertial first responder localisation in large-scale indoor training environments.

Type of content
Theses / Dissertations
Publisher's DOI/URI
Thesis discipline
Human Interface Technology
Degree name
Doctor of Philosophy
Publisher
Journal Title
Journal ISSN
Volume Title
Language
English
Date
2021
Authors
Khan, Humayun
Abstract

Accurately and reliably determining the position and heading of first responders undertaking training exercises can provide valuable insights into their situational awareness and give a larger context to the decisions made. Measuring first responder movement, however, requires an accurate and portable localisation system. Training exercises of- ten take place in large-scale indoor environments with limited power infrastructure to support localisation. Indoor positioning technologies that use radio or sound waves for localisation require an extensive network of transmitters or receivers to be installed within the environment to ensure reliable coverage. These technologies also need power sources to operate, making their use impractical for this application. Inertial sensors are infrastructure independent, low cost, and low power positioning devices which are attached to the person or object being tracked, but their localisation accuracy deteriorates over long-term tracking due to intrinsic biases and sensor noise. This thesis investigates how inertial sensor tracking can be improved by providing correction from a visual sensor that uses passive infrastructure (fiducial markers) to calculate accurate position and heading values. Even though using a visual sensor increase the accuracy of the localisation system, combining them with inertial sensors is not trivial, especially when mounted on different parts of the human body and going through different motion dynamics. Additionally, visual sensors have higher energy consumption, requiring more batteries to be carried by the first responder. This thesis presents a novel sensor fusion approach by loosely coupling visual and inertial sensors to create a positioning system that accurately localises walking humans in largescale indoor environments. Experimental evaluation of the devised localisation system indicates sub-metre accuracy for a 250m long indoor trajectory. The thesis also proposes two methods to improve the energy efficiency of the localisation system. The first is a distance-based error correction approach which uses distance estimation from the foot-mounted inertial sensor to reduce the number of corrections required from the visual sensor. Results indicate a 70% decrease in energy consumption while maintaining submetre localisation accuracy. The second method is a motion type adaptive error correction approach, which uses the human walking motion type (forward, backward, or sideways) as an input to further optimise the energy efficiency of the localisation system by modulating the operation of the visual sensor. Results of this approach indicate a 25% reduction in the number of corrections required to keep submetre localisation accuracy. Overall, this thesis advances the state of the art by providing a sensor fusion solution for long-term submetre accurate localisation and methods to reduce the energy consumption, making it more practical for use in first responder training exercises.

Description
Citation
Keywords
Ngā upoko tukutuku/Māori subject headings
ANZSRC fields of research
Rights
All Right Reserved