A new facial recognition system could be capable of identifying matches based on multiple stills from poor quality video captured at different time points.
It could also reconstruct faces in 3D, according to a research team at the Queensland University of Technology (QUT).
Lead investigator Professor Fookes of QUT said there were a number of challenges to working in an unconstrained environment, where people were not necessarily aware their image was being captured.
‘The idea is to identify people who are walking around naturally in a public space, for example, at major sporting events, concerts, in airports, train stations and so on,’ he said. ‘So we are often working with images that may have low lighting or shadows, where the face may not be clearly visible all of the time and where the resolution may be very small.’
Development of such a surveillance system would be of benefit to law enforcement agencies, which were often hampered in their investigations by poor quality vision and images.
The focus of the research project is to develop mathematical algorithms that make it possible to take features from video and convert that into a model capable of recognising and matching facial features.
The ultimate aim is to use multiple cameras in space to reconstruct a face in 3D, or use multiple images over time of the same face to reconstruct into 3D.
‘We want to get as much information from as many sources as we have available, so if that is 10 cameras then we will use information from 10 cameras,’ said Fookes.
‘Once we have the information, the system will then be able to identify a shortlist of possible candidates and it will then be up to a human observer to authenticate the correct match.’