International research collaborations

Led by Professor Ralph R Martin with our partner in China, Tsinghua University, where our own Visual Computing Group has been cooperating and collaborating with the Computer Graphics Group for over 20 years (and published more than 60 joint papers). Professor Martin is a guest professor at the university.

One of his projects set out to explore new approaches to comparing, classifying and searching visual media – and meaningful ways to edit and resynthesise visual media. Another (funded with a 1.6m Chinese Ministry of Science and Technology investment) looked at the processing of digital geometry, images and video sequences. It aimed to promote understanding of shape, image and video using neural networks and other ideas based on the structure of the human visual system, and to develop mathematical tools necessary for processing and which would allow computers to learn about things they see.

Led by Dr Hantao Liu with our partner in Italy, the Polytechnic University of Milan, the research focused on the need to build safe systems for autonomous vehicles (AVs) in the future. We know that AVs will need to co-exist and share the road with human-driven vehicles and that some AVs might be partially/conditionally autonomous (so a human driver will be in control some of the time).

This research looked at the challenge of ensuring AVs make safe and appropriate driving decisions in any traffic situation by making them reactive and able to customise their decisions. It aimed to gather data from smart road infrastructure on features of the ‘other road users’ based on characteristics that typically signify age, gender and driving experience and then build models that allowed for variables in human behaviour presented by those human drivers rather than interacting on a ‘one-size-fits-all’ attention and reaction model.

Led by Dr Bailin Deng with our partner in The Netherlands, Utrecht University, this research set out to make architectural and industrial design easy and efficient. It looks at the constraints on 3D design (such as building facades from flat glass panels, a custom cog in a machine, or a new roof on an existing building) and the design tools that might be slow and difficult to work with. The research set out to alleviate this difficulty by introducing a new design paradigm: working in increasing resolution levels. For instance, designing a facade with 500 glass panels in the desired approximate shape, and then refining it to 200 panels. This aims to give designers intuitive control in the design process (allowing them to refine details of a coarse shape) while offering constrained shape optimisation without any redundant effort.

Led by Professor Paul Rosin with our partner in Canada, Carleton University this project set out to simplify the complexity of the computer algorithms involved in the comparison of shapes, which forms the basis of exploration and retrieval systems. Several technologies, including video games, special effects, and architectural visualization, rely on the use of 3D shapes for the creation of more complex 3D content. In this context, the ability to easily search for specific shapes is important (the available repositories of shapes have become so large nowadays that manual search is impractical).

However, because many of the 3D models available are not tagged with meta-data (such as keywords describing the objects that they represent), an important task that can be accomplished by computer algorithms is to search for specific shapes according to their content. Professor Rosin’s team knew that while his can be carried out by the user providing an initial shape or several shapes as examples of the desired objects, and then retrieving similar objects from the datasets, finding matching shapes is challenging since many objects are articulated (e.g. a human’s limbs can rotate) or can undergo non-rigid deformation (e.g. the bending of an elephant’s trunk). The goal of this research is to facilitate the exploration of large repositories of shapes and retrieval of models from these repositories with the use of “canonical forms”. These canonical forms will deform the objects to standardize their appearance, thereby simplifying the complexity of the computer algorithms involved in the comparison of shapes, which forms the basis of exploration and retrieval systems.

Led by Dr Richard Booth with our international partner in France, Université Paris 8, this research look at how, in decision making contexts (medical diagnosis, security, environmental policies), it is important systems can deal with large amounts of knowledge in a robust way.

While artificial Intelligence allows computers to represent and reason with the data they collect, the goal of this project is to develop systems that are able to understand their data and make inferences from it, even in imperfect, "messy" contexts in which the information is uncertain or incomplete. This requires both machine-readable data format, or language, in which to represent the computer's knowledge of the world and of the various concepts occurring in it. The team used description logics to meet this challenge. But they also needed to establish inference mechanism allowing the computer to extract new knowledge that was perhaps only implicit in the data so that the project-developed systems would deliver this, even in the imperfect world of the web.

Dr Nervo Verdezoto Dias with our international partner in South Africa, University of Cape Town

Digital interventions to support maternal and child health (MCH) are widespread. Yet the impact of these interventions in low-income communities is still limited. We believe this is, in part, due to the top-down nature of digital health development and propose to take a participatory and community-centred approach to this domain. We are forming a multidisciplinary, cross-cultural, and cross-geographical consortium of researchers, technology designers, healthcare professionals and community stakeholders, policy makers, and grassroots citizens' organizations to explore the potential of information and communication technologies (ICTs) to enhance maternal and child health and wellbeing during the antenatal and postnatal period in South Africa.