Indicative Data Science

3-Dimensional (3D) models of cities are beneficial or even essential for many applications, including urban planning/development, energy demand/consumption modelling, emergency evacuation and responses, lighting simulation, cadastre and land use modelling, flight simulation, positioning and navigation (particularly for autonomous cars in urban canyons and disabled users requiring accessibility), and Building Information Modelling (BIM). Despite the importance of the 3D models, they are not available or being updated frequently for many areas/cities. This can be due to the process of generating and updating (by current technologies such as LiDAR (Light Detection and Ranging)) being computationally and financially expensive, time-consuming, and requiring frequent updates due to the dynamic nature of cities. This fellowship will propose and implement a crowdsourcing-based approach to create accurate 3D models from the free to use and globally available data of Global Navigation Satellite Systems (GNSS). The effects of urban features, such as buildings and trees, on GNSS signals, i.e. signal blockage and obstruction, and attenuation, will help to recognise the shape, size, and materials of urban features, through the application of statistical, machine learning (ML) and artificial intelligence (AI) techniques. The use of freely accessible raw GNSS data, which can be accessed on any current Android device, will enable the production of up to date 3D models at no or low cost, of particular value in developing regions where these models are not currently available.

GNSS is the most widely used positioning technique because of free-to-use, privacy-preserving, and globally available signals. However, GNSS signals can be blocked, reflected and/or attenuated by objects, e.g. trees, buildings, walls and windows. While blockage, attenuation and reflection of GNSS signals are common in urban canyons and indoors, making the positioning unreliable, inaccurate or impossible, the affected received signals can act as an indicator of the structure of the surrounding environments. This means, for example, if the signals are blocked or attenuated, then the size and shape of the obstacles or the type of media/material the signals have gone through or been reflected by can be understood. This needs the precise locations of satellites, and the receiver, and also predicted signal strength level at each location and time. The crowdsource-based framework, i.e. a mobile app for data capture and a web mapping application for upload of GNSS raw data, will allow the project to have well-distributed data both in space and time. This will ultimately lead to higher quality (more spatially and temporally accurate, complete, precise) 3D models. However due to the complexity of data, as neither the receiving mobile devices nor the broadcasting satellites are fixed, some novel data mining techniques, based on already existing statistical, ML, and AI techniques, need to be developed during this fellowship. They will handle the high volume, the velocity of change, and the complexity of the spatio-temporal GNSS raw data with high levels of veracity. The spatio-temporal patterns will be used for creating and updating the 3D models of cities at a high level of detail (LoDs), i.e. approximating the façade and the building materials, e.g. windows, from which the signals are reflected or have gone through. The 3D models will feed into 3D-mapping aided GNSS positioning (and integrated with other signals e.g. WiFi) which can ultimately provide more continuous and accurate GNSS positioning in urban canyons and indoors.