Rapid 3D face modelling explored for contactless facemask fitting
Computer vision experts at the University of Southampton have proposed using camera sensors to generate a 3D model of the face in a quick and efficient facemask fitting system.
A research team led by Electronics and Computer Science's Dr Hansung Kim are developing a prototype that virtually fits masks to faces to ensure an adequate seal for wearers such as frontline healthcare workers.
The project is one of a number of innovative activities across south coast universities that have been funded by the Wessex Academic Health Science Centre to address the wide-ranging challenges of COVID-19.
Surgical masks are designed to prevent external splashes but poor fitting can still allow contaminated air to pass through gaps and into the lungs.
Current fit tests spray a flavoured mist over the wearer within a hood, with the tester tasting the spray if the mask is not suitably shaped and leaks. Most tests take between 15 and 20 minutes, with any negative results needing to then be retested on another day.
The method results in many respirators being disposed and wasted, with recent studies finding that several mask models fail fit tests for between 20% and 60% of people.
The new Southampton approach would hugely speed up the process, optimise user protection, reduce the number of trained people needed to lead the tests - freeing up clinical staff - and save precious resources such as respirators, sprays and cleaning solutions.
Hansung, an Associate Professor in the Vision, Learning and Control Research Group, says: Generating 3D human face models from captured images can virtually fit masks to faces to choose the best mask for individual users. However, current 3D laser scanners are bulky and expensive, while cheaper sensors such as mobile phone cameras have been shown to have high construction errors.
"We propose a system that utilises two small and low-cost colour and depth cameras to provide accurate depth as well as colour. The set up time would be less than five minutes to generate 3D face models in real-time, with the user able to test multiple face models with various facial expressions. Our software would then automatically find the best fitting mask by calculating mesh-fitting errors between the 3D face and mask models."
The Southampton team aims in time to build the system as a single package that can be used without any setup or calibration.
Following the initial prototype, Hansung is interested in exploring collaboration and partnership opportunities with business and industry to further the concept of customised facial mask production and modified PPE based upon human face models.