|M.Sc Student||Issawy Elias|
|Subject||Deep Learning Hybrid Model for Elevation|
Calculation on Smartphones
|Department||Department of Civil and Environmental Engineering||Supervisor||Dr. Sagi Dalyot|
|Full Thesis text|
Perhaps some of the most critical features in smartphones are the ones related to localization, since they serve as the major geospatial part in most location-based services and applications. Although the 2D (horizontal) localization usually delivers reliable results, which for most services suffice, the 3rd dimension - the elevation - still suffers from reduced accuracy and reliability. This is mostly because of the technological and physical limitations of the embedded smartphone sensors, with environmental conditions affecting the reliability and accuracy of observations.
This research investigates analytical ways and methods, using Artificial Neural Network (ANN) in deep learning, to improve the calculation of the elevation by integrating readings from various sensors embedded in smartphones, such as barometer and gravimeter, supplemented with data from external mapping and environmental infrastructures, such as topographic and weather databases. The motivation is to develop a hybrid computational model that improves the elevation observation received on smartphones, with the focus on urban and concealed (e.g., trees canopy) areas. To this end, various field experiments were carried out to study the different effects of environmental conditions on the sensors, with a survey of the available accurate external sources that can be used. A comprehensive field survey was conducted, collecting large volumes of data used in the experiments. The analyses and evaluations of the developed hybrid model show very promising results of elevation calculation by the different embedded smartphone sensors, validating the hybrid model robustness and reliability in different conditions and locations.