Visual Sweden City Plattform

A digital twin of a city is an important strategic resource that can play a significant role in the quality of our future built environment. Digital twins can contribute to equipping our cities for a faster and more dynamic societal development and for more complex, digital, and data-driven planning processes. Rich data models create opportunities for cross-disciplinary analyses where parallel questions are analyzed simultaneously, such as social and technical issues, reducing risks of "roadblocks" in the planning process and leading to an increased democratization of planning and decision-making. In the digital twin, a large number of different scenarios can also be simulated in a way that is not possible in the physical world or where data is analyzed individually.

To create a digital twin of a city, a very large amount of data is required, as well as effective data collection methods. The project addresses the problem of how to feed a digital twin with data and at the same time visualize it with high visual accuracy. The project aims to explore a wide range of data collection methods from satellites to vehicle-borne scanning and to greatly automate this process through state-of-the-art techniques in machine learning, synthesis, image analysis, and 3D reconstruction. The goal is to generate a digital twin with semantic information and metadata with minimal manual effort while increasing mapping capability.

The project's various stakeholders work together with geodata in different forms and there is a common need to map and analyze geodata to improve services, methods, and to develop society. The challenges faced by Swedish cities are also largely common, including those related to climate change, demographics, quality of life, planning of societal development, or changing transportation systems. This makes it necessary to develop methods and methodologies that are general and can be applied across a wide range of areas and multiple needs. The methods must also be cost-effective and agile so that new questions and data can be added and adapted to handle a wide physical scale, from a single location to an entire region or surroundings. The project's stakeholders reflect a range in applications and applicability regarding physical scale and data collection methods, from ground perspectives to regional levels, from laser to satellite, from driving simulators (VTI) to disaster management (MAXAR). They all share the dependence on creating a geographic digital twin and making it rich in metadata while having a realistic visual representation that supports communication, both between professional actors and citizens (Norrköping municipality, Malmö city, Linköping municipality).

The project aims to investigate how different collection methods can be used to reinforce each other in creating digital twins. How machine learning (ML) through deep neural networks, synthesis and image analysis can be used to enrich the final result and automate parts of the mapping and 3D reconstruction process. The project also aims to investigate and develop methods where synthetic data is used to train neural networks that can be applied in the physical reality.

This could make it possible for more types of objects or classes to be mapped, even if there is currently little data available for traditional ML training of these. This method has been successfully used in medical technology research at Linköping University, and dissertations have indicated great potential in a urban context as well. The use of synthetic training data for mapping cities is considered state of the art today.

A key part is also to support further national and international research in the fields of machine learning, image analysis and visualization by making available a unique dataset of over a range of Swedish cities and regions.

The questions and scope for applications of image analysis and 3D reconstruction of cities are worldwide, so the target audience is the international research arena as well as developing service companies.