CAMERA PhD Opportunities
We are excited to be able to advertise four PhD opportunities. The successful students will be part CAMERA, which performs world-leading multi-disciplinary research in Intelligent Visual and Interactive Technology.
Closing date : Sunday, March 28, 2021
Department of Computer Science
Training and Rehabilitation in VR
VR holds great promise for improving the way athletes train and the way people with injuries rehabilitate. This project aims to create interactive VR simulations that allow users to train specific motor skills; for example, a tennis player may work on her serve or a person with a shoulder injury may work on extending the arm at the shoulder joint. Based on related work we hypothesise that, by allowing a user to experience a motor skill in VR, the user will be more likely to achieve their performance goals.
In this project, we will explore methods of creating interactive experiences based on motion capture, machine learning (ML), VR design and Human-Computer Interaction (HCI). We will test the new methods with athletes and people affected by injuries, working together with researchers from Sports & Exercise Science and Health.
Immersive VR Video Creation from 360° Videos
Virtual reality photography and video is a rapidly evolving field within visual computing (see Richardt et al. 2020 for a survey). To feel truly immersed in virtual reality, one needs to be able to freely look around within a virtual environment and see it from the viewpoints of one’s own eyes. Full immersion requires that viewers see the correct views of an environment at all times. As viewers move their heads, the objects they see should move relative to each other, with different speeds depending on their distance to the viewer. This is called motion parallax and is a vital depth cue for the human visual system that is entirely missing from existing 360° VR video.
The goal of this project is to capture the real world with a single 360° video camera and to recreate its appearance for new, previously unseen views, to enable more immersive virtual reality video experiences. To do this, the project aims to develop new monocular depth estimation techniques for reconstructing dynamic scene geometry from 360° video input, and novel-view synthesis techniques that can produce high-quality, temporally coherent, time-varying VR video of dynamic real-world environments. Particularly important are the convincing reconstruction of visual dynamics, such as moving people, cars and trees. This experience will provide improved motion parallax and depth perception to the viewer (like Bertel et al., 2020) to ensure unparalleled realism and immersion.
Department for Health
On-court workload assessment of tennis players using markerless motion analysis
The ability to accurately track on-court player movement is becoming increasingly relevant in tennis, enabling the objective quantification of player workload, which can be utilised in both performance and commercial (e.g. media statistics) applications. To date, automatic player tracking has received some attention in the literature, with several computer vision algorithms proposed (e.g. Archana & Geetha, 2015; Bloom & Bradley, 2003). However, no system has been fully validated to our knowledge and therefore there exists concerns over their accuracy for certain applications. Specifically, these approaches typically estimate the centre of mass of the athlete as the centroid of their silhouette, which is not representative of the player’s centre of mass in many poses. This will introduce a degree of error in the quantification of each movement effort, which will clearly become a much larger problem when movements across a whole training session or match (typically > 200 sprint accelerations even in youth players; Pereira et al., 2016) are of interest.
A further problem of workload monitoring systems currently adopted is the arbitrary thresholds typically used to define movement intensities (e.g. >5.5 m/s movement would be classed as sprinting for every player), which clearly do not capture the metabolic cost of an individual’s movement. To accurately track the individual workloads and make inferences about the likely fatigue an athlete is experiencing, it is paramount that such intensity domains are individualised. An automatic system capable of providing the individualised work capacity metrics would considerably improve workload monitoring of tennis players. In order to overcome the aforementioned challenges, the proposed project would validate markerless player tracking methods against gold standard marker-based motion capture and utilise these technologies alongside work capacity tests to produce an individualised workload assessment tool.
Modelling and prediction of spinal injury in rugby
Spinal catastrophic injury in rugby activities (e.g. scrums and tackles) are not very common but can be very impairing (e.g. tetraplegia, or quadriplegia) and generate high societal costs. Governing bodies, such as the World of Rugby, have successfully minimised such injuries during scrummaging by changing the game rules. These changes were informed by biomechanical analysis carried out at University of Bath. Unfortunately, rugby tackling still remains a very risky part of the game and a cause-effect relationship between tackling technique and resulting injuries is still lacking.
The aim of this projects is to unveil such relationship and causally link specific tackling situations to injury diagnoses. This will be achieved by using probabilistic, generative machine learning to represent the probabilistic relationship between qualitative and quantitative tackling features and injury diagnoses. This is a great opportunity to extend the state-of-the-art in machine learning whilst working directly with domain experts to provide real-world insights and benefits in the fields of bio-mechanics and sports injury prevention.
Qualitative data will include description of injurious events gathered via interviews with injured players, whilst quantitative data will be retrieved from the simulation of the same injuries in a computational framework using musculoskeletal models of the spine. The simulation framework will be firstly validated against imaging data (e.g. x-rays and MRIs) of real-world spinal injuries, and then used to create a set of ‘synthetic’ data of theoretical injurious scenarios. The ‘synthetic’ datasets will include quantitative information of vertebrae motion, and forces applied to the spine during injurious events, which is key to understand the injury mechanisms and compare the simulated injuries with real injury diagnoses and medical images. The final step will consist in the creation of the probabilistic model that will use as input the qualitative descriptors of tackling scenarios and quantitative simulation results, and will provide the most likely injury diagnosis.
Funding is available to candidates who qualify for Home fee status. Following the UK’s departure from the European Union, the rules governing fee status have changed and, therefore, candidates from the EU/EEA are advised to check their eligibility before applying.