Health Technology PhD: Using Computer Vision to Study Everyday Movement and Fall Risk in Older Adults

Health Technology PhD: Using Computer Vision to Study Everyday Movement and Fall Risk in Older Adults

Quick Decision

This Exeter project combines computer vision, biomechanics and clinical falls research. It will use markerless pose estimation to derive movement-stability measures from video, validate them against motion capture, compare movement patterns in people with and without a fall history, test when instability can be detected, and explore low-cost on-device processing. The starting point is a simulated-home dataset pairing video, motion capture and wearables, with about 225 participants expected by the studentship start. The project asks for a strong quantitative/computing degree and Python or MATLAB; a Master’s is described as ideal, while prior clinical experience is not required. [S01, S02]

The key question is what “predict falls” means in this study. Movement-quality measurement, near-fall recognition, short-horizon warning and prediction of an individual’s future fall risk require different labels, timelines and validation. Event counts, prediction horizon, data splits, external testing and privacy arrangements are not yet specified. Because NICE advises against using falls-risk prediction tools to predict an individual’s risk, applicants should clarify the intended endpoint and clinical use. A controlled simulated home can support method development but cannot by itself establish performance in ordinary homes. [S01, S08, S09]

Funding and application dates also need written confirmation. The DLA provides up to 3.5 years of funding, tuition and £5,000 total RTSG. Official pages conflict on whether the main deadline is 8 or 12 December 2026, whether references are due 8 January or 15 February 2027, and whether £21,805 is a 2026/27 or 2027/28 rate. Plan against 8 December at noon and 8 January for references until Exeter confirms otherwise. [S02–S05]

Research Problem and Doctoral Work

Falls are common in later life and often occur during ordinary tasks rather than a short clinic test. The project description reports that about one-third of people aged 65 and over fall each year and cites around 210,000 emergency admissions a year in England. A separate public-health indicator reports 221,600 emergency hospital admissions due to falls among people aged 65 and over in England in 2024/25. These figures establish a substantial public-health problem; they do not imply that any single camera or model can prevent falls.

The advertised study uses the STEPPING OUT Home Scene, a simulated home at Exeter’s VSimulators facility within Professor Sallie Lamb’s ERC-funded programme. Participants stand up and sit down, turn, move around furniture, open doors, change surfaces, and respond to slips and trips. Multi-camera video, motion capture and wearable sensors record movement; participants’ clinical and falls histories are known, including people who have and have not fallen. Collection is ongoing, with around 225 participants expected by the start of the studentship. The project document does not give the number of fall events, near-falls or participants in each health subgroup, and 225 is a participant target rather than a count of falls.

The four advertised work areas are:

1. Use markerless pose estimation to derive stability and movement-quality measures, then validate them against motion capture.
2. Find movement patterns and instability markers that distinguish fallers from non-fallers, including near-falls and moments before loss of balance.
3. Test predictive performance and how early movement changes can be detected.
4. Adapt the methods for low-cost devices used in homes, care homes and hospitals, with processing on the device.

The project names pose estimation, deep learning for movement sequences and self-supervised learning. Interpretability is a stated priority so clinical users can understand and act on results. Proposed deliverables include a validated analysis pipeline, open benchmarks for video-based stability assessment, evidence on prediction from ordinary video, and a tested low-cost prototype. No camera configuration, pose model, prediction window, hardware specification, compute budget, latency target, data split or external validation cohort is stated. Treat the deliverables as project aims, not guaranteed outcomes.

A useful methodological distinction to settle early is the target label. Movement-quality assessment can be validated against motion capture. Near-fall or loss-of-balance recognition needs an operational event definition and timing annotation. Pre-impact detection asks whether a model can warn during an ongoing fall sequence, usually over a short horizon. Future fall-risk prediction asks whether a person will fall over a longer follow-up period and needs suitable prospective outcomes. These labels cannot be substituted for each other. In this dataset, participant history and controlled perturbations may support movement and instability analyses, but the description does not explain whether prospective everyday falls will be observed.

A scoping review of markerless motion capture found 39 eligible studies covering 3,114 participants (mean age 75.8); gait speed, stride length and step width were recurring measures, while methods and reporting varied and data-loss reporting needed improvement. Another review of 243 sensing studies found only 21 with real-world or mixed real-world validation in older adults; most in that subgroup addressed post-fall detection, while pre-impact prediction was a minority. This supports the project’s focus on validation and interpretable measures, but also highlights the gap between controlled demonstrations and reliable use in homes.

NICE NG249, published in April 2025, states that falls-risk prediction tools should not be used to predict an individual’s risk of falling. Its rationale cites very-low-quality evidence and inadequate sensitivity or specificity for existing tools. This is guidance for clinical assessment, not a prohibition on research into video-derived movement or balance signals. It does mean the applicant should ask how the intended endpoint and clinical claim will differ from a standalone risk score, and what action a clinician or care provider could take based on a result.

Potential analyses to discuss—not additional advertised requirements—include participant-level rather than frame-level splits to prevent leakage between recordings of the same person; calibration and false-alert burden alongside AUC or accuracy; robustness to viewpoint, occlusion, lighting and assistive devices; and a held-out setting or cohort before making claims about home deployment. The project has not confirmed whether these are already planned.

School, Programme and Field Context

The University of Exeter is the degree-awarding institution named by the official project and award pages. The project PDF names the potential route as “PhD in Public Health and Sport Sciences,” the Department of Public Health and Sport Sciences, and the Richards Building and Medical School at St Luke’s Campus. Exeter’s current Public Health course page instead labels the route MPhil/PhD Public Health and describes the department as an interdisciplinary public-health hub. Confirm the formal programme and registration title on the application or offer. VSimulators, where the project’s simulated-home data are collected, is a separate research facility at Exeter Science Park; it is not the stated study location or a second degree-awarding institution.

Exeter’s subject-specific REF 2021 results provide relevant context: the university’s UoA24 submission in Sport and Exercise Sciences, Leisure and Tourism received 79% 4-star, 20% 3-star and 1% 2-star ratings. The Public Health course page reports 11th in the UK for research power in REF 2021 UoA2 (Public Health, Health Services and Primary Care). These are department- or unit-level signals, not a project ranking and not evidence of a particular computer-vision group’s performance.

The project PDF says training will span computer vision, data science, biomechanics and clinical falls research. The Public Health course page describes regular supervisory meetings and interdisciplinary research, but does not specify this studentship’s teaching, modules, compute allocation or training schedule. Exeter lists a four-year maximum for full-time PhDs and says funded students should plan thesis submission by the funding end date; the DLA itself funds up to 3.5 years. Confirm the expected thesis timeline and how the final six months are supported if needed.

Supervisor and Lab Assessment

The official project PDF names Dr Naser Taleshi as lead supervisor, with Professor Sallie Lamb, Professor Xujiong Ye and Associate Professor Dominic Farris. Their combined research areas plausibly connect movement biomechanics, falls and ageing, health research and computer vision. The PDF does not allocate their supervisory roles or specify meeting frequency. Public Health programme information says regular meetings are typical, but an applicant should confirm the arrangements for this award.

Recent team work provides relevant building blocks, with important limits:

• A 2026 preprint by Taleshi and Exeter collaborators analyses narratives for 270 confirmed real-world falls; 211 cases remained after excluding reports without enough detail. Object interactions and everyday transitions feature in the resulting framework. This motivates studying furniture and task context, but the article is not peer reviewed, uses retrospective narratives, and does not validate camera-based prediction.
• A 2025 Scientific Reports paper by Taleshi, Lamb and collaborators studies balance responses on a dynamic platform using a model-predictive-control framework. It offers biomechanical context for perturbation and stability measures; it is not an older-adult video prediction study.
• A 2024 Nature Medicine perspective by Lamb, Williams and Young discusses falls science and prevention. It is useful clinical framing, not a new intervention or prediction-system trial.
• A 2024 markerless movement-analysis paper co-authored by project co-supervisor Xujiong Ye develops motion biomarkers for knee disorders. It shows adjacent experience in camera-based movement analysis, but the population and outcome differ from older-adult falls.
• A 2026 preliminary digital motor-cognitive assessment paper co-authored by Taleshi reports a small 26-person concussion comparison. It shows activity in digitally captured movement assessment, but is not direct evidence for older-adult fall prediction.

The project’s named facility, VSimulators, combines virtual reality, a motion platform, biomedical sensors and motion capture for controlled human-environment research. The project PDF separately confirms that the STEPPING OUT scene uses multi-camera video, motion capture and wearable sensors. The facility’s general capabilities do not establish the specific camera set-up, student access schedule or deployment hardware.

Funding, Application, and Living Cost

Item	Official wording	Planning implication
Application deadline	6008 award-page body: 12 December 2026. The award summary and DLA scheme page say midday 8 December 2026.	Use 8 December at noon as the safe date; obtain written confirmation.
References	6008 page: 8 January 2027. Central DLA information: 15 February 2027.	Ask referees to submit by 8 January unless Exeter confirms otherwise. Missing documents can prevent shortlisting.
Interviews	6008 page: likely 15–24 February 2027. Central page also contains stale wording for interviews/outcomes in 2026.	Request the current interview timetable and format.
Stipend	£21,805 is described as the current 2026/27 rate in the page body and central DLA page; the award summary labels the same amount 2027/28. UKRI lists £21,805 as its 2026/27 minimum.	Treat £21,805 as the current baseline only; confirm the indexed amount for 2027/28.
Award	Up to 3.5 years full-time stipend and tuition, plus £5,000 total Research Training Support Grant over 3.5 years.	Confirm award start/end dates, indexation and the budget’s allowable uses.
International eligibility	International applicants may compete for a limited number of full awards, up to 30% of the cohort. Exeter says successful 2027 international DLA awardees receive a fee waiver for the international/home difference.	Eligibility is competitive and residency rules are complex; obtain a written fee-status and award confirmation. Visa, healthcare surcharge and moving costs are not included.

At the current stipend baseline, £21,805 divided across 12 months is about £1,817 per month. Exeter’s April 2026 student budget estimate is £1,300–£2,000 per month. The current monthly equivalent is roughly £517 above the lower estimate but about £183 below the upper estimate, before one-off arrival costs or visa and healthcare charges. The university estimate varies with accommodation and lifestyle; use an actual accommodation quote and confirm the 2027/28 stipend before deciding whether the package is workable.

The project PDF directs applicants to use the application link in that project document. The central DLA asks for a personal statement, CV, full transcript(s), two references and English evidence when applicable; documents must be in English, with certified translations of qualifications. Applicants can apply to up to three DLA projects, submitting a separate application for each and naming their preferred project in the statement. The scheme is competitive. General DLA guidance asks for a UK First or Upper Second (or overseas equivalent) in an appropriate science/technology subject; a Lower Second may be considered with a Master’s or significant relevant non-academic experience. That is separate from the project PDF’s “strong first degree” wording, whose grade threshold is not defined there.

Applicant Fit and Preparation

Required background and skills

The project PDF lists a strong first degree in computer science, data science, artificial intelligence, engineering, applied mathematics, physics, biomechanics or a related area; Python or MATLAB programming; good written and spoken communication; and ability to work in an interdisciplinary team. The PDF says a Master’s is “ideally” held, so it is presented as desirable preparation rather than a strict condition. The general DLA academic threshold still applies. Ask the team how it interprets “strong first degree” alongside the DLA’s First/Upper Second rule and Lower Second exception. Prior clinical or health-sector experience is explicitly not required.

Desirable experience and suggested preparation

The project lists experience with images, video or signals, or machine learning as desirable. The project description does not require a particular pose-estimation library, deep-learning framework or prior clinical qualification. As preparation, applicants could bring a small reproducible computer-vision or time-series project, demonstrate a careful validation design, and show how they communicate model limitations to non-technical collaborators. These are suggestions based on the advertised work, not additional entry criteria.

The work may be less appealing to applicants seeking a purely theoretical AI doctorate, a purely clinical falls-prevention study, or an already specified product-development role: it combines methodological development with older-adult movement data and currently leaves several modelling and deployment choices open.

Career Paths and Market Context

Possible routes include academic research in movement science and ageing, digital health, clinical research analytics, health technology and medical-device development. The project may build transferable skills in motion analysis, computer vision and health data, but moving into clinical deployment may also require stronger experience in software engineering, privacy governance, clinical validation and medical-device regulation. Skills England’s life-sciences assessment and DSIT’s AI Labour Market Survey describe broad sector growth and demand for data, AI and regulatory skills; neither estimates vacancies for fall-prediction specialists or graduate outcomes from this project. These sources provide adjacent skills context, not a placement rate or employment guarantee. A portfolio of validated code, robust model evaluation, clear documentation, responsible health-data handling and clinical collaboration may strengthen a candidate’s profile. Skills England life-sciences assessment · DSIT AI Labour Market Survey 2025 · UK Life Sciences Jobs Plan

Next Actions and Questions to Verify

Before applying

1. Email the PGR Admissions Team and Dr Taleshi to confirm the application cutoff (8 or 12 December), reference deadline (8 January or 15 February), and the required project-specific application route. Use the earlier dates until they reply.
2. Ask whether “strong first degree” means a specific classification and how the project’s “ideally a Master’s” wording interacts with the general DLA criteria, especially for overseas equivalents or Lower Second applicants.
3. Confirm international fee-status eligibility, the number of international award places, and whether the fee waiver covers the full published fee difference for this specific offer.
4. Verify whether the 2027/28 stipend is £21,805 or another indexed amount, and whether the programme registration is “PhD in Public Health and Sport Sciences” or “MPhil/PhD Public Health.”

At interview

• What is the primary endpoint: movement quality, near-falls, short-term loss-of-balance prediction, or future falls over follow-up? What are the event definitions and prediction horizon?
• How many people and events are expected in faller, non-faller, neurological and near-fall groups? How will class imbalance and participant-level data leakage be handled?
• Will evaluation include motion-capture agreement, calibration, sensitivity/specificity, false alarms per unit time, lead time, external validation or a prospective cohort?
• What low-cost devices, on-device compute, latency and privacy constraints define a successful prototype? Who controls the output and what action should follow it?
• How will the intended clinical use align with NICE NG249? What are the plans for consent, ethics, raw-video access and retention, de-identification, security, and open benchmark release?
• Which supervisor leads computer vision, data analysis, clinical interpretation, biomechanics and deployment? What training, group meetings and access to VSimulators are included?

Before accepting an offer

Obtain the final stipend schedule, fee classification and coverage, RTSG rules, award end date, expected thesis submission date, and formal programme title in writing. Budget for accommodation, transport, deposits, visa, immigration health surcharge and moving costs separately; do not assume the stipend’s current-year label applies to 2027/28.

Evidence and Sources

Project, award and eligibility

• Project description PDF: computer vision and falls in older adults
• EPSRC Doctoral Landscape Award, reference 6008
• Central DLA eligibility, funding and application information
• EPSRC DLA scheme page
• UKRI studentship minimum stipend guidance

Supervisors, research group and recent papers

• Naser Taleshi: research outputs
• Real-world falls framework analysis (2026 preprint)
• Balance control on a dynamic platform (2025)
• Digital cognitive–motor assessment for concussion (2026)
• The science of falls and how to prevent them (2024)
• Markerless motion analysis in knee disorders (2024)
• Applied Cognitive Science at Exeter
• VITAL research group
• VSimulators facility

Field, programme, public health and career context

• Markerless motion capture scoping review (2025)
• Fall detection and pre-impact prediction scoping review (2026)
• NICE NG249 recommendations
• NICE rationale for falls-risk prediction guidance
• Exeter Public Health MPhil/PhD programme
• Exeter research-degree durations
• Exeter REF 2021 results by subject
• Exeter student living-cost estimates
• OHID Fingertips falls-admission indicator
• ICO guidance on special-category data
• ICO research provisions and special-category processing
• Skills England life-sciences skills assessment (2026)
• AI Labour Market Survey 2025
• UK Life Sciences Jobs Plan