Review of 11 sensors and 2 software platforms used in our anxiety detection research — eye tracking, heart rate, skin conductance, video, and motion capture. Includes specs, comparisons, and sample data.
 Pupil Labs Core Head-mounted eye tracker Dark pupil + 3D model · 5-point calibration 1080p @30 Hz · 720p @60 Hz · 480p @120 Hz |
 Tobii Pro Glasses 3 Wearable eye tracker 100 Hz gaze sampling |
 SMI Eye Tracking Systems Research-grade eye tracker 60 Hz sampling |
 Polar H10+ Chest strap — HR, HRV, IBI 64 MHz microprocessor · ECG sensor |
 Moofit HW401 Wearable heart rate monitor 64 MHz microprocessor · ECG sensor |
 Empatica E4 Wristband Wrist wearable — EDA, BVP, temp, accel Multi-sensor (EDA + BVP + temp + accel) |
 TEA CAPTIV T-SENS GSR Skin conductance (wireless) 32 Hz sampling · 20 g · 8 hr battery |
 BioPac EDA Sensors Electrodermal activity measurement |
 AXIS P1275 Camera Network surveillance camera HDTV 1080p · WDR-Forensic capture |
 AXIS P1245 Camera Compact network camera HDTV 1080p |
 OptiTrack (Slim X13) Motion capture system |
 OpenFace Open-source facial analysis Landmark detection · Head pose · Action unit recognition · Gaze estimation |
 Noldus FaceReader Commercial facial analysis 468-point face model · 20 Action Units · 99% accuracy on ADFES |
From Experiment Scenario.pdf — a 5-phase, 30-minute session:
| Phase | Duration | Description |
|---|---|---|
| 1. Interview | 10 min | Semi-structured demographic interview, informed consent |
| 2. Sensor Setup | 5 min | Fit Pupil Labs Core, Polar H10+, and Moofit HW401; calibrate |
| 3. Baseline | — | Reading task to get resting-state measurements |
| 4. Psychometric Testing | 15 min | HADS, STAI, BFI-10, and FQ on screen while sensors record |
| 5. Debrief | — | Participant feedback; mental health resources provided if needed |
From Threshold.pdf — literature-based thresholds used for detecting anxiety:
| Sensor | Measure | Anxiety Threshold | Source |
|---|---|---|---|
| Pupil Labs Core | Fixation duration | < 250 ms | Laeng et al. (2012) |
| Pupil Labs Core | Saccade peak velocity | > 500 deg/s (for ≥ 15° saccades) | Di Stasi et al. (2013); van der Lans et al. (2013) |
| OpenFace | Brow furrowing (AU 4) | ≥ 3.0 on 0–5 intensity scale (FACS C) | Ekman & Friesen (1978); Gavrilescu & Vizireanu (2019) |
| OpenFace | Lip tightening (AU 24) | ≥ 3.0 on 0–5 intensity scale (FACS C) | Ekman & Friesen (1978); Gavrilescu & Vizireanu (2019) |
| TEA GSR | Skin conductance response | > 0.05 µS | Boucsein (1992) |
| Polar H10+ / Moofit | RMSSD | < 50 ms | ESC/NASPE Task Force (1996) |
| Polar H10+ / Moofit | SDNN | < 50 ms | ESC/NASPE Task Force (1996) |
| Document | What's in it |
|---|---|
List of Sensors.pdf |
Specs and capabilities for each sensor |
Sensor Comparison.pdf |
Side-by-side comparison |
Sensors (Eye-tracking, HRV, GSR, Camera).pdf |
Overview by measurement type |
OpenFace vs Noldus.pdf |
Facial analysis software comparison |
Experiment Scenario.pdf |
Experimental protocol |
Threshold.pdf |
Anxiety detection thresholds with references |
Sensor Documentation.pdf |
Consolidated sensor reference documentation |
The data/ folder has CSVs from psychometric testing sessions (2024-06-13 to 2024-06-24). Full column definitions are in the data dictionary.
| File | Rows | Description |
|---|---|---|
eye_metrics.csv |
264 | Per-question eye-tracking summary — pupil dilation & blink rates aggregated per question across all 3 sessions. Not heart-rate variability (formerly misnamed HRV.csv). See DATA_DICTIONARY.md for column details. |
hr.csv |
4032 | Heart rate from multiple sensors (Polar H10+, Moofit) with confidence scores |
ibi.csv |
2394 | Inter-beat interval series |
Psychometric_Test_Results.csv |
88 | Question-level responses (HADS, STAI-S, STAI-T, FQ, BFI) with timestamps |
sed.csv |
34171 | Raw eye tracking — head position, gaze direction, pupil size, eye openness |
sed_fix.csv |
34171 | Processed eye tracking — adds gaze difference, fixation detection, fixation duration |
pip install -e . # the sensor_data package + runtime deps
# or reproduce the exact CI-pinned environment:
pip install -r requirements.txt
# fully reproducible, hash-locked install (all extras):
pip install -r requirements.lockSee analysis/explore_data.ipynb for a walkthrough that loads each CSV and plots heart rate, pupil dilation, fixation durations, and (illustrative-only) IBI-derived RMSSD/SDNN — which the notebook flags as not valid HRV — from the sample session.
- Picking sensors for a multimodal physiology study
- Comparing what each device can actually do
- Designing experiments with multiple concurrent sensors
- Understanding the data formats each platform outputs
The data/ folder holds sample human physiological and psychometric recordings from a single consenting participant, collected during anxiety-detection sensor evaluation. The participant gave written informed consent, including consent to share the pseudonymised sample openly for research and educational use. No direct identifiers are included (no names, contact details, dates of birth, or device identifiers), and the data are handled as special-category personal data under the GDPR and the French Data Protection Act. No IRB number applies; governance rests on that data-protection framework plus explicit consent. Intended for research and educational use only; do not attempt to re-identify the participant. Full statement: DATA_ETHICS.md.
- Multimodal-Multisensor — Longitudinal study using these sensors
- CalmSense — Stress detection system built on multimodal signals
Pupil Labs Core · Polar H10+ · Moofit HW401 · TEA CAPTIV T-SENS · Empatica E4 · OpenFace · Noldus FaceReader · OptiTrack · AXIS Cameras
Biometric Sensors · Eye Tracking · Heart Rate Variability · Galvanic Skin Response · Electrodermal Activity · Anxiety Detection · Psychometric Testing · Multimodal Sensing · OpenFace · Pupil Labs
If you use this resource, please cite:
Bose, U. (2026). Sensor: A Review of Sensors and Software Platforms for Anxiety-Detection Research.
GitHub. https://github.com/urme-b/Sensor
@misc{bose2026sensor,
author = {Bose, Urme},
title = {Sensor: A Review of Sensors and Software Platforms for Anxiety-Detection Research},
year = {2026},
url = {https://github.com/urme-b/Sensor},
note = {Review of 11 sensors and 2 software platforms for multimodal anxiety detection}
}Tagged releases are archived on Zenodo for long-term preservation and a citable DOI; the DOI badge will be added here once the first release is archived (deposit metadata lives in .zenodo.json).
This repository is dual-licensed by content type:
- Source code — the analysis notebook,
sensor_datamodule, and tests: MIT - Sample data & documentation — the CSVs, data dictionary, PDFs, and this README: CC BY 4.0
- Product images (
images/) — property of their respective manufacturers, included for identification/review only and not covered by CC BY 4.0; seeimages/CREDITS.md
When reusing the data or documentation, please retain attribution (see Citation).