Professor Sean Gong, Professor of Visual Computation at the School of Electronic Engineering and Computer Science and Queen Mary Turing Fellow, was recently awarded the Institute of Engineering and Technology (IET) 2020 Achievement Medal for Vision Engineering.
Professor Sean Gong
The IET Achievement Awards exist to recognise individuals from all over the world who have made exceptional contributions to the advancement of engineering, technology and science in any sector.
Professor Gong is a pioneer and world-leading researcher on visual surveillance based on computer vision, artificial intelligence, machine learning and video analysis. His innovative research is being employed daily in forensic analysis and surveillance for security and crime prevention.
Commenting on the award, Professor Gong said: “It was a pleasant surprise to receive the award and be globally recognised for my contributions to the field of Vision Engineering. I’m incredibly honoured and humbled to be named alongside a host of other leading engineers throughout the engineering and technology industries.”
Since the late 1980s, Professor Gong has been working to develop systems that can be used to track and identify public space activities of people and vehicles in unstructured large scale video data, and in particular re-identify (RE-ID) their whereabouts over different locations and time.
These AI powered RE-ID systems allow users to search for people when facial recognition is not feasible, for example due to poor lighting conditions, in low-resolution imagery, or when individuals are obstructed or looking away from view.
“Whilst there were existing tools using AI from computer vision and machine learning to detect and track individuals, they weren’t always sufficient for the scale of some tasks where you might have to search tens of thousands of hours of video footage, covering large pools of the population from different physical places,” said Professor Gong.
“Our specific technique for the re-identification of people can search and match in high speed, more than 100 times faster than real-time, making tasks that might have taken existing systems 2-3 years to complete possible in just a few days. It also provides a way to identify individuals when visual information is not accessible for all sorts of legal, operational or ethical reasons.”
The innovative solutions developed by Professor Gong and his research team have seen them collaborate with Industry and Government partners worldwide, helping to fight crime, enable more efficient and safer transport, and employ cleaner and more user friendly urban environment designs in emerging smart cities.
In 2007 Professor Gong founded spin out company Vision Semantics Ltd, based on his research. More recently, he has been working with commercial partners to identify innovative ways to use the underlying RE-ID technology for a number of applications outside of the public safety and security sectors, including E-commerce.
Professor Gong said: “As this technology fundamentally works by matching visual objects in different environments over space and time, the same approaches can be applied to smart shopping. For example, if you took at a low-resolution photo on your phone of someone walking down the street wearing an item of clothing you liked, you could then use a similar technology to find that item of clothing in an online shop, despite the photo looking very different from the high quality image taken in a studio. This type of approach allows us to begin to link customer’s online purchases with their offline behaviours.”
Current AI and deep learning techniques are dependent on two main assumptions, access to huge amounts of data and infinite computer power. However, it is likely that access to both will be limited in future.
“One of the areas of research we’re focusing on at the moment is looking at how we can learn from decentralised and distributed smaller datasets,” said Professor Gong. “With increasing data protection legislations, we will need to begin to shift existing centralised big data deep learning paradigms to adjust to data on a more local level, for example from multiple countries, or cities."
"We need to respond to this new environment, developing AI technologies that will provide similar levels of information as we would get from big global data sets but are more in line with user-ownership of data, and the user-centred, localised learning that will be required going forwards.”