In-Bed Human Pose Classification Using Sparse Inertial Signals

A simulation-based study of twelve sleeping poses, using synthetic inertial data and machine learning to address scarce labelled examples.
Sleep posture
Machine learning
Synthetic data
Speaker
Event / host

Omar Elnaggar

AI 2020 · SGAI 40th International Conference on Artificial Intelligence · Medical and Legal Application Stream · Online

Event date

Thursday, the 17th of December, 2020

Abstract
Recognising detailed body postures during sleep is challenging when few labelled wearable measurements are available. This talk presents a classification approach for twelve benchmark sleeping poses and a way to generate synthetic training data. Computer-graphics simulation provides a controlled setting for exploring the method, including its response to measurement noise. An ensemble of support-vector-machine classifiers connects the generated data with posture recognition. The work establishes an in-silico proof of concept and motivates subsequent evaluation with wearable sensors and human participants.

Cover slide: In-Bed Human Pose Classification Using Sparse Inertial Signals.

Work with Frans Coenen and Paolo Paoletti.

Related research: Sleep wearables.

← Back to Talks

Back to top