Introduction
Monitoring the physiological and emotional state of passengers during long-haul flights is a growing area of interest for improving cabin comfort and safety. However, traditional clinical monitoring—utilizing "wet" electrodes and restrictive wiring—is unsuitable for the passenger experience. While unobtrusive sensors like capacitive electrodes or textile-integrated devices offer a solution, they are notoriously susceptible to motion artifacts. This paper presents a multimodal sensor system integrated into an airplane seat that utilizes sensor fusion to mitigate noise and provide high-fidelity health data.
System Architecture and Sensor Integration
The researchers developed a "smart" seat prototype equipped with two categories of sensors: those measuring physiological signals and those measuring environmental artifacts.
The physiological suite includes:
- Contactless Capacitive ECG: Embedded in the backrest to measure heart rate through clothing.
- Textile Respiration Sensor: A resistive sensor integrated into the seatbelt.
- Finger Electrodes: Dry sensors for Electrodermal Activity (EDA) and secondary ECG.
- Skin Temperature: Measured at the fingertips.
To address the noise inherent in these sensors, the system incorporates artifact-detecting hardware: pressure sensors in the backrest to monitor body shifts and 3-axis accelerometers on the fingers to track hand movement.
Methodology: Sensor Fusion and Artifact Mitigation
The core innovation lies in the dual-strategy approach to data processing. For heart rate monitoring, the system uses artifact gating. By monitoring the auxiliary pressure and acceleration channels, the system can identify specific time windows where the passenger is moving.
As shown in the figure above, when the contact pressure fluctuates significantly (indicating a shift in posture), the corresponding contactless ECG signal becomes heavily corrupted. The system automatically identifies these "Regions of Artifacts" and discards them, ensuring that only high-quality data is used for R-peak detection.
For emotion recognition, where discarding data is less desirable, the authors employed ensemble classifiers. Instead of a single model, separate classifiers were trained for each modality (ECG, EDA, Respiration). A fusion layer then combines these outputs via majority or confidence voting. This allows the system to maintain predictive power even if one or more sensors are temporarily offline due to noise.
Experimental Results
The system was evaluated with 20 participants across two studies focusing on daily activities and emotional response.
- Heart Rate Detection: Without the artifact removal logic, the capacitive ECG achieved only a 40% accuracy in R-peak detection compared to ground-truth wet electrodes. By applying the sensor fusion gating, accuracy improved to 88%.
- Emotion Classification: The ensemble approach proved robust against data loss. Even when 53% of the data was corrupted by artifacts, the system could still classify emotional states (amusement, anger, contentment, sadness, neutral) with an accuracy improvement of up to 14% over single-modality methods.
- Operational Feasibility: During typical "calm" flight activities like sleeping or reading, the system provided nearly 3 hours of clean data over a simulated 12-hour period.
Discussion and Future Work
While the system excels during periods of low activity, it remains sensitive to high-movement phases such as eating or working, where nearly 100% of the data may be discarded. Future iterations will focus on improving the robustness of the capacitive sensors to varying clothing thicknesses and investigating the impact of real-world cabin vibrations. Nevertheless, this study demonstrates that by treating noise as a measurable variable, unobtrusive sensors can provide clinical-grade insights in non-clinical environments.
Summary updated