Proceedings of the ACM on Human-Computer Interaction (PACM HCI), Volume 9, Issue 5 • 2025
iBreath: Usage of Breathing Gestures as Means of Interactions
Highlights
- Breathing gestures detected via bio-impedance with event-level accuracy >93% and time-step accuracy >95%
- Gestures studied: single, double, triple clicks and SOS; single click preferred, triple click disliked
- Users learn new gestures in ~50 seconds (five guided trials); median gesture durations 3.5–5.3 s
- User-dependent model (14 training trials, ~140 s) achieves >90% precision/recall; user-independent >85%
- Hardware: AFE AD5941 + nRF52840, 100 kHz, 50 mV stimulus, 20 Hz sampling; two electrodes under armpits
- Post-processing with low-pass, front-follows-back, and majority-rule strategies improves predictions
- Cost-effective prototype (~USD 40), Bluetooth streaming; magnitude channel more informative than phase
- Design guidelines address electrodes, sensing parameters, model choice, and interaction design
- Robust across sitting, lying, and walking; data augmentation improves user-independent generalization
- Potential applications include assistive tech, sterile or discreet environments, and wearable control
Abstract
iBreath is a wearable system that detects discrete breathing gestures for hands-free interaction using upper-body bio-impedance sensing. The system models changes in impedance caused by lung volume variations and recognizes gestures such as single, double, triple clicks and a custom SOS through a lightweight neural network with data augmentation and post-processing. Two lab studies (n=34) evaluate robustness and user experience across sitting, lying, and walking. Time-step accuracy exceeds 95% and event-level accuracy exceeds 93% in cross-validation. Users learn new gestures quickly (about 50 seconds for five guided trials) and report low mental, physical, and temporal demand. User-dependent models trained with 14 trials reach over 90% accuracy, while user-independent models trained on data from 21 participants exceed 85%. Participants prefer single-click gestures and dislike triple-clicks; median gesture durations are 3.5–5.3 seconds. The paper contributes a low-cost hardware prototype (~$40), a gesture recognition pipeline, evaluations, and eight design guidelines for bio-impedance-based breathing interaction.
Click to read Full Paper.