Research Article

Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion

Number: Advanced Online Publication Early Pub Date: August 1, 2026
EN TR

Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion

Abstract

Behavioral biometric identification with inertial measurement units (IMUs) has largely relied on sensors placed on the lower body or wrists, implicitly treating the task as gait recognition. This study investigates a complementary, head-centric configuration in which three IMU clusters (NXP FXOS8700CQ accelerometer and FXAS21002C gyroscope, Teensy Prop Shield design) are rigidly co-mounted on a safety helmet at the crown (IMU0) and at left and right ear level over the mastoid region (IMU1 and IMU2). Raw measurements are combined with pairwise differential (Δ) and summation (Σ) cross-IMU channels to represent both common-mode head motion and spatially heterogeneous residual motion. Twelve classifiers spanning feature-based, recurrent, attention-based, and convolutional paradigms were evaluated on data from 48 participants. The evaluation used non-overlapping 60 s groups, stratified group 10-fold cross-validation, and fold-specific scaling and feature selection to control temporal and preprocessing leakage. A residual 1D CNN with squeeze-and-excitation channel attention achieved 99.46 ± 0.5 % accuracy (macro-F1 = 99.47 %) at the 4 s window level and perfect segment-level classification under the adopted protocol. It significantly outperformed every non-convolutional baseline (Wilcoxon signed-rank, p = 0.002), whereas the tested convolutional variants were statistically indistinguishable. Ablation experiments clarify what the system requires. No tested CNN channel subset differed significantly from the full 54-channel representation after correction for multiple comparisons: a single IMU was sufficient (99.35–99.51 % versus 99.48 %), and crown and ear-level placements were indistinguishable. Identity was recoverable from either signal component: Σ-only achieved 99.36 %, while Δ-only achieved 99.18 % with the CNN but only 79.45 % with a Random Forest on the same 18 channels. The differential residual therefore contains substantial identity information, although the present experiments do not establish its physical mechanism or show that it is independent of gait. A 1 s observation produced 98.53 % accuracy, with performance saturating near 4 s. In the three participants with repeated recordings, holding out an entire recording collected on another day and in another environment reduced accuracy by only 0.24 percentage points (Δ-only: 0.18 points), providing preliminary evidence of cross-session stability. The same architecture achieved 99.27 ± 1.0 % accuracy on the independent HelmetPoser dataset using a single six-channel IMU. Finally, the full, uncompressed 54-channel network was exported to a dependency-free C++ implementation and evaluated on a Teensy 4.1 (Cortex-M7, 600 MHz). The implementation required 558 ms per 4 s window, corresponding to a 13.9% duty cycle, and used 792 KB of flash and 269 KB of RAM. These results position helmet-mounted inertial sensing as a practical and underexplored modality for closed-set wearable identification and verification.

Keywords

References

  1. Connor, P., & Ross, A. (2018). Biometric recognition by gait: A survey of modalities and features. Computer Vision and Image Understanding, 167, 1–27.
  2. Thang, H. M., Viet, V. Q., Thuc, N. D., & Choi, D. (2012). Gait identification using accelerometer on mobile phone. Proceedings of the 2012 International Conference on Control, Automation and Information Sciences (ICCAIS), 344–348. https://doi.org/10.1109/ICCAIS.2012.6466615
  3. Woodman, O. J. (2007). An introduction to inertial navigation. University of Cambridge, UCAM-CL-TR-696.
  4. Dehzangi, O., Taherisadr, M., & ChangalVala, R. (2017). IMU-based gait recognition using convolutional neural networks and multi-sensor fusion. Sensors, 17(12), 2735. https://doi.org/10.3390/s17122735
  5. Deb, S., Yang, Y. O., Chua, M. C. H., & Tian, J. (2020). Gait identification using a new time-warped similarity metric based on smartphone inertial signals. Journal of Ambient Intelligence and Humanized Computing, 11(10), 4221–4230. DOI: 10.1007/s12652-019-01659-7
  6. Andersson, R., Bermejo-García, J., Agujetas, R., Cronhjort, M., & Chilo, J. (2024). Smartphone IMU sensors for human identification through hip joint angle analysis. Sensors, 24(15), 4769. https://doi.org/10.3390/s24154769
  7. Abdulrahman, L. S., Sabir, A. T., & Maghdid, H. S. (2026). A biometric dataset for unconditioned gait identification using onboard smartphone sensors. Frontiers in Computer Science, 8, 1752141. DOI: 10.3389/fcomp.2026.1752141
  8. Semwal, V. B., Gaud, N., Lalwani, P., Bijalwan, V., & Alok, A. K. (2022). Pattern identification of different human joints for different human walking styles using inertial measurement unit (IMU) sensor. Artificial Intelligence Review, 55(2), 1149–1169. DOI: 10.1007/s10462-021-09979-x

Details

Primary Language

English

Subjects

Neural Networks

Journal Section

Research Article

Early Pub Date

August 1, 2026

Publication Date

-

Submission Date

July 20, 2026

Acceptance Date

July 31, 2026

Published in Issue

Year 2026 Number: Advanced Online Publication

APA
Turan, R. Y., & Tektaş, M. (2026). Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion. Electronic Letters on Science and Engineering, Advanced Online Publication. https://izlik.org/JA43MK95WC
AMA
1.Turan RY, Tektaş M. Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion. Electronic Letters on Science and Engineering. 2026;(Advanced Online Publication). https://izlik.org/JA43MK95WC
Chicago
Turan, Recep Yavuz, and Mehmet Tektaş. 2026. “Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion”. Electronic Letters on Science and Engineering, no. Advanced Online Publication. https://izlik.org/JA43MK95WC.
EndNote
Turan RY, Tektaş M (August 1, 2026) Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion. Electronic Letters on Science and Engineering Advanced Online Publication
IEEE
[1]R. Y. Turan and M. Tektaş, “Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion”, Electronic Letters on Science and Engineering, no. Advanced Online Publication, Aug. 2026, [Online]. Available: https://izlik.org/JA43MK95WC
ISNAD
Turan, Recep Yavuz - Tektaş, Mehmet. “Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion”. Electronic Letters on Science and Engineering. Advanced Online Publication (August 1, 2026). https://izlik.org/JA43MK95WC.
JAMA
1.Turan RY, Tektaş M. Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion. Electronic Letters on Science and Engineering. 2026. Available at https://izlik.org/JA43MK95WC.
MLA
Turan, Recep Yavuz, and Mehmet Tektaş. “Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion”. Electronic Letters on Science and Engineering, no. Advanced Online Publication, Aug. 2026, https://izlik.org/JA43MK95WC.
Vancouver
1.Recep Yavuz Turan, Mehmet Tektaş. Head-Centric Inertial Biometrics: Learning Human Identity from Helmet-Mounted Motion. Electronic Letters on Science and Engineering [Internet]. 2026 Aug. 1;(Advanced Online Publication). Available from: https://izlik.org/JA43MK95WC