Machine Learning and Deep Learning models to discern indoor from outdoor environments based on data recorded by a tri-axial digital magnetic sensor

Wearable devices (e.g., smartphones, smartwatches, etc.) are increasingly used for position estimation of users both in indoor environments to provide personalized contents (e.g., malls, museums, crowded venues, etc.) and outdoor environments to provide additional information about their way of moving. Existing mainly solutions exploit the use of GPS which can provide very accurate location information in outdoor environments, while in indoor environments due to a variety of physical barriers that can attenuate the GPS signal (e.g., walls, floors) the quality of the provided information can be very poor. Other approaches which consist in the use of Wi-Fi, Bluetooth, barometers, and light sensors have been investigated and presented in the literature. Despite these approaches can provide very accurate indoor location information, their use is limited because they require the presence of beacons (i.e., Wi-Fi, Bluetooth) and suitable mapping surveys of the interested areas (i.e., barometer, light sensors) which is not feasible for outdoor environments. A technology that's been around for an extremely long time which however, in this thesis, has an innovative and very interesting application is the magnetometer. Thanks to its intrinsic properties, a magnetometer can measure variation of the Earth’s magnetic field strength result of the disturbances due to the presence of ferromagnetic materials which are easier to find in indoor environments (e.g., ferromagnetic materials, electric power lines, etc.) with respect to the outdoor ones. In addition, the advantages of the use of a magnetometer are that i) it is already integrated in most wearable devices; ii) it is less-power consuming with respect to the abovementioned technologies; and, finally, iii) it does not require the use of additional infrastructure (i.e., beacons) and/or areas mapping. Therefore, the aim of this thesis is to apply machine learning techniques to discern indoor from outdoor environments by looking at the local magnetic field strength variations recorded by a magnetometer during activities of daily living of 20 subjects.