Aim: To develop a highly precise model which determines whether a mobile phone user is indoors or outdoors. A combination of mobile phone sensors could be exploited, such as light, magnetometer/barometer, accelerometer, gyrometer and perhaps overall signal from nearby Wi-Fi points. However, no location information can be inferred, to respect the privacy of users.
Background: Digital contact tracing apps can help break the chain of coronavirus infections and save lives by complementing manual tracing. The virus does not stop at borders, and therefore the European Commission and Member States are working to ensure the seamless cross-border interoperability of these apps. Right now, these contact tracing apps cannot consider whether an exposed user to COVID-19 has been exposed indoor vs. outdoor, which is an important factor in order to calculate the overall risk to this user.
A high-accuracy model, which combines sensory information from modern mobile phones to estimate whether the user is indoor or outdoor. Current state-of-art results present an accuracy of 85%, but we need an accuracy of 90%+ in order to convince Google/Apple to include this indoor-outdoor aspect in the risk calculation of their Bluetooth GAEN API.
Type of work expected:
- This work includes mostly data analysis.
- Probabilistic modelling algorithmic knowledge and understanding is important (Random Forests, SVM, Deep Learning – CNN).
This project will be in co-operation with Pervasive Systems and CYENS Centre of Excellence, which is the developer of the official COVID-19 contact tracing app for the country of Cyprus, named CovTracer-EN.
Andreas Kamilaris (firstname.lastname@example.org)