Metrically-accurate 3D Face Shape Estimation for Applications in Health Care

This project aims to develop techniques for metrically-accurate 3D face reconstruction from uncalibrated, commodity cameras. Existing computer vision approaches such as multi-view stereo and 3D morphable models will be combined with the latest advances in deep learning to achieve this aim. This fills a gap between existing solutions - they fall between the extremes of inaccurate single-image reconstruction and expensive professional performance capture. Both extremes are limited by their performance in an unconstrained setting. Initial results have shown great promise for deep learning techniques to perform well in this domain. This project would build upon existing work to allow for inexpensive, accurate reconstruction which performs well in an unconstrained setting.

The applications of such techniques range from increasing immersion in VR to personalised implant design. The project will focus on medical applications, particularly those in orthodontics. This takes advantage of the strong connections between the Department and the School of Dentistry at the University of Leeds. The proposed approach will allow for cheaper care which is accessible to a wider group of people, improving clinical outcomes. The project complements existing research at the Department and has the potential to greatly improve the cost and accuracy of facial reconstruction for a broad range of applications.