Generating Digital Twins of Roads

This research aims to provide the needed semantic segmentation methods for the reality capture of digital twins' foundation data enhancing 3D reconstruction, asset certainty and overall
spatial context and understanding. Spatial context offers the ability for digital twins to integrate asset data with local area knowledge provided by existing national data holdings (e.g. OS MasterMap) contributing to enhanced resilience. This project can significantly strengthen the level of maturity (complexity and connectedness) of digital twins enabling the next generation of smart applications, such as performance optimisation by real time data capture, failure prediction and future scenario modelling. Tools exist that can automatically annotate visual data using deep learning architectures. Yet the outcome is not immediately usable in our context. The generation of geometric Digital Twins requires the presence and processing of both visual and spatial data to derive sufficient meaning and generate the richest possible outcome. The challenge in this context is how to automatically annotate combined visual and spatial datasets with a sufficient degree of detail (small and large assets, asset defects, etc.) to meet road network inspection guidelines.