Human Crowdsourcing Data for Indoor Location Applied to Ambient Assisted Living Scenarios

• Main Author: Santos, Ricardo Bruno Barbeiro dos
• Publication Date: 2018
• Format: Master thesis
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: http://hdl.handle.net/10362/56377
• topic: Ambient Assisted Living; Crowdsourcing; Indoor Location; Fingerprinting; Time Series Similarities; Machine Learning; Domínio/Área Científica::Engenharia e Tecnologia::Outras Engenharias e Tecnologias

In the last decades, the rise of life expectancy has accelerated the demand for new technological solutions to provide a longer life with improved quality. One of the major areas of the Ambient Assisted Living aims to monitor the elderly location indoors. For this purpose, indoor positioning systems are valuable tools and can be classified depending on the need of a supporting infrastructure. Infrastructure-based systems require the investment on expensive equipment and existing infrastructure-free systems, although rely on the pervasively available characteristics of the buildings, present some limitations regarding the extensive process of acquiring and maintaining fingerprints, the maps that store the environmental characteristics to be used in the localisation phase. These problems hinder indoor positioning systems to be deployed in most scenarios. To overcome these limitations, an algorithm for the automatic construction of indoor floor plans and environmental fingerprints is proposed. With the use of crowdsourcing techniques, where the extensiveness of a task is reduced with the help of a large undefined group of users, the algorithm relies on the combination ofmultiple sources of information, collected in a non-annotated way by common smartphones. The crowdsourced data is composed by inertial sensors, responsible for estimating the users’ trajectories, Wi-Fi radio and magnetic field signals. Wi-Fi radio data is used to cluster the trajectories into smaller groups, each corresponding to specific areas of the building. Distance metrics applied to magnetic field signals are used to identify geomagnetic similarities between different users’ trajectories. The building’s floor plan is then automatically created, which results in fingerprints labelled with physical locations. Experimental results show that the proposed algorithm achieved comparable floor plan and fingerprints to those acquired manually, allowing the conclusion that is possible to automate the setup process of infrastructure-free systems. With these results, this solution can be applied in any fingerprinting-based indoor positioning system.