Fuzzy proximity estimation by opportunistic use of Wi-Fi and GSM fingerprints

Abstract

Lack of communication is often the main problem during emergencies inside buildings. People at the back of a crowd will push on those at the front resulting in people at the front being crushed. Currently used techniques like cameras to observe people and guidance signs, are not efficient enough for people's safety. This motivated the choice for this FYP to try and make it easier to guide people during an emergency or evacuation. To reach this goal, sensors available on mobile phone devices, due to the ever increasing popularity of these devices that people carry them with them all the time, will be made use of to extract context information like location of a person and proximity to other people. With this information guidance to detected groups can be provided to be guided to safety by sending messages to the people in the group (Group Addressed Messaging). The sensors in question are Wi-Fi and GSM signals. Localisation and proximity are the main concepts on which this thesis is based. To find the location of a particular user, a Fingerprint Match Algorithm will be implemented using the K NN machine learning algorithm. GSM and Wi-Fi signal information will be gathered as fingerprints from the user's mobile devices. For the Fingerprint Match Algorithm to work it needs a set of initial examples to be able to locate unseen fingerprint measurements. These examples will be collected using a mobile application which gathers fingerprints and send them to the central server to be stored in a database. The user then uses a different application to measure fingerprints on the fly and these will also be sent to the central server were the Fingerprint Match Algorithm is used to compare them with the examples in the database. This way the location can be found. Having found each user's location then proximity between users can be estimated and extract group formations and structures to be known for emergency cases.