Computer vision and image analysis uses UCL Computer Science's deep expertise in machine learning algorithms to make sense of the huge amounts of data found in pictures and video. Being able to interpret visual data means that machines can now ‘see’ new ways to assist humans across an increasingly broad spectrum of activities.
What can computer vision help us with? We’re still finding new answers to that question. We’re applying computer vision to wild animal conservation, robotic surgeries, medical imaging, augmented reality environments for training or entertainment – even solving inverse problems in mathematics and bioinformatics.
In short, computational vision and imaging research represents such a paradigm shift, human imagination is still catching up.
What are computer vision and imaging researchers working on at UCL Computer Science?
3D modelling and depth estimation
Multimodal foundation models such as GPT-4o, Gemini and Dall-E have already processed vast amounts of 2D image data. UCL Computer Science researchers are developing depth estimation algorithms to build on these foundations, so these models can begin to make inferences about the 3D environments these pictures represent.
Research in this area has already been developed by our academics into tools and platforms that are used by millions of people every day. Startups co-founded by our academics include AI video platform Synthesia (recently valued at $2bn) and Matrix Mill (recently acquired by Niantic, the creators of Pokemon Go).
Our leading computer vision researchers often work across both academic and industry settings, ensuring UCL Computer Science staff and students retain access to the very latest in both pure and applied research.
Medical image computing and surgical intervention
At UCL’s newly established Hawkes Institute, advanced medical imaging analysis is giving clinicians unprecedented new insights into how and why diseases occur and develop. These medical image computing tools can be applied across the body to investigate different illnesses, from Alzheimer’s to prostate cancer.
Computer vision is also providing clinicians with the tools to take action. Image-guided surgical interventions give surgeons a better view of hard-to-reach areas during operations, helping them make better decisions during diagnoses and operations.
This research has already been translated from bench to bedside, through spinouts such as Odin Vision, an AI-driven endoscopy application that was recently acquired by Olympus.
Solving inverse problems in MRI and PET scans
Arising in almost all fields of science, inverse problems are solved by working backwards, using what you can observe to find hidden causes. Machine learning and computer vision are essential tools for this, using statistical inference to predict solutions to these problems.
Our computer vision researchers are using machine learning with inverse problem-solving approaches to increase the accuracy of MRI scans for the head and lungs, helping clinicians more effectively identify and diagnose potential health issues.
Biodiversity monitoring
Working in collaboration with the UCL Centre for Biodiversity and Environment Research, UCL computer scientists are trialling new ‘human in the loop AI’ approaches, using volunteers to train neural networks to autonomously identify different bat species.
This interdisciplinary collaboration has also developed new self-supervised learning algorithms to more easily classify different species captured on film by remote wildlife monitoring systems.
What's coming next for computer vision and imaging research at UCL Computer Science?
Foundational machine learning models present enormous advantages for speeding up innovation. Researchers can build rapidly on the immense capabilities and data resources of these models, creating teachable software platforms that can observe and anticipate interactions between objects in 3D spaces.
Eventually, these systems will learn to codify the rules of physics that govern the way objects and people move relative to their surroundings. Once this is achieved, it will be possible for machines to join us safely and productively in our many different physical environments.
