Advancing Digital Twin Creation for Smart Cities: Innovative Approaches in Visual Representation and Simulation
Abstract
The United Nations (UN) estimates that by 2050, 70% of the world’s population will reside in urban areas. Given the increased demand for services, there is a need to create sustainable, well-managed, and resilient cities. Authorities collect vast amounts of data about cities and their citizens to achieve this. The ability to analyze, understand, and extract knowledge from urban data is essential if cities are to remain sustainable and become resilient as they expand across multiple dimensions (e.g., physical size, economic activity, and population). Recently, the increase in sensors and Internet-of-Things (IoT) infrastructure in cities has enabled new forms of data to be collected more regularly, often in real-time. City planners are investigating ways to monitor urban progress and climate change, simulate patterns in a city, and assist with urban modeling and planning. This has led to the generation of "Smart City Digital Twins", next-generation sensors and communications technology-based systems within a physical urban domain. This thesis presents a comprehensive exploration of Digital Twins (DTs) for smart cities, focusing on four critical aspects fundamental to the automatic generation of smart city digital twins: scene generation for creating digital replicas based on textual feedback, DTs for lighting analysis in urban areas, 3D reconstruction for modeling and simulation, and facial anonymization to address security and privacy concerns in DTs. Urban planners and city administrators can employ automated DTs to create accurate virtual replicas of physical entities, enabling detailed analysis, simulation, and decision-making across various domains.
This thesis investigates scene generation for DTs, particularly in controlled image generation from scene graphs and layouts. In this context, it examines how textual input can facilitate the creation of 2D images and 3D models for automatic model generation in a DT sandbox. It introduces a standard methodology for evaluating the performance of image generation models, comparing scene graph-based and layout-based methodologies. The experimental analysis of the Visual Genome and COCO-Stuff datasets reveals that the layout-based image generation models surpass scene graph-based methods in complex scenarios, offering valuable insights into the future development of DT technologies based on user text-based feedback. In addition, this thesis examines the role of DTs in lighting analysis within urban contexts. It reviews advanced works on using DTs for simulating and visualizing various lighting scenarios, considering factors such as technology readiness levels of lighting simulation suites, urban lighting planning, and walkability and wildlife assessment in smart cities. The findings underscore the potential of DTs in optimizing urban lighting systems and contributing to sustainable urban development. This thesis then proposes a Digital Twin Authoring Tool (DTAT) that automates the creation of virtual replicas of vehicles by exploring state-of-theart 3D reconstruction methods from a smart mobility perspective in smart cities. Utilizing pre-trained models, the DTAT processes data from IoT sensors, enabling the reconstruction of 3D models that serve as DTs in virtual environments. These digital replicas are integrated into a Virtual Reality (VR)-based environment, allowing stakeholders to conduct simulations that optimize transport systems and enhance smart mobility. Lastly, this thesis addresses the critical issue of privacy preservation in smart cities through facial anonymization. As visual data proliferation escalates, protecting individuals’ privacy becomes paramount in smart city environments. This ethical consideration is a crucial aspect of DT development. This thesis proposes a machine learning-based face-anonymization method to address this, demonstrating its efficacy in real-time surveillance for smart city digital twins. The proposed approach ensures that the quality and intelligibility of data for DTs are preserved while safeguarding personal privacy. The findings contribute to the advancement of DTs in smart cities, offering novel methodologies and tools for developing a DT sandbox for smart city applications. These contributions are instrumental in realizing the full potential of DTs, facilitating the development of more intelligent, efficient, and sustainable urban environments.