Wearable sleep posture classification using one-shot learning of biomechanical signatures

Recognising complex sleep postures from wearable joint measurements, with one-shot learning and an examination of the challenges beyond controlled experiments.
Wearable sensing
Sleep posture
Machine learning
Speaker
Event / host

Omar Elnaggar

BioMedEng26 · University of Liverpool, Liverpool, UK

Event date

Friday, the 4th of September, 2026

Abstract
How can wearable sensors distinguish sleep postures that share the same broad body position but differ in the configuration of the limbs? This talk explores wrist and ankle orientations as interpretable signatures of twelve complex postures. A one-shot learning approach creates additional training examples from a single measured example of each posture. Evaluation moves from computer-generated posture data to controlled participant measurements and more realistic indoor conditions. The discussion considers both the promise of learning from limited data and the practical difficulties posed by sensor drift and similar postures, setting out the questions that remain before dependable long-term monitoring.

Cover slide: Wearable sleep posture classification using one-shot learning of biomechanical signatures.

Work with Roselina Arelhi, Frans Coenen, Andrew Hopkinson and Paolo Paoletti.

Oral presentation, session O26: Craniofacial, Soft-Tissue and Sports Biomechanics. Scheduled for 15:20–15:30 in Space Six.

Related research: Sleep wearables.

← Back to Talks

Back to top