Research

HomeProjectsResearchAboutConnect Me

Co-author · IEEE T-BIOM · 2026

Keystroke-Aligned Body Motion Patterns for Short-Burst Continuous Smartphone Authentication: A Proof-of-Concept Study Using Motion Capture

N. Cariello, L. Nguyen, R. Gallagher, I. Kurtzer, K. S. Balagani, P. Gasti

IEEE Transactions on Biometrics, Behavior, and Identity Science · doi:10.1109/TBIOM.2026.3674400

Read it on IEEE Xplore

Continuous authentication using keystroke dynamics works well over long windows — 20 to 120 seconds, and 5 to 10 in recent work. But a great deal of real phone use is far shorter: typing a URL, entering a search query, firing off a one-line reply. Those bursts last one or two seconds, which is where the state of the art stops being usable.

Equal error rate over a 1-second window

Keystroke dynamics alone

9.5–11.9%

With phone and body motion

1.5%

What we did

Keystroke dynamics combined with laboratory-grade 3D motion capture and smartphone motion, using keystroke events as temporal anchors for feature extraction — extending earlier work that paired body motion with swipe-based biometrics.

Findings

  • A 1.5% equal error rate over 1-second authentication windows when all available features are used.
  • Keystroke-only approaches reach 9.5% to 11.9% EER under the same conditions, so adding phone and body motion cuts the error several times over.
  • Evaluated on a publicly available dataset of 42 users.
  • Identifies which body regions and feature types carry the most signal in short windows, as a roadmap for doing this with phone sensors and wearables rather than a motion capture lab.

Keywords

Behavioral biometrics
Continuous authentication
Keystroke dynamics
Motion capture
Multimodal authentication
Smartphone security

Worked in

Python
NumPy
Pandas
Matplotlib
scikit-learn
Jupyter

Research Assistant · IEEE T-BIOM · 2024

Posture and Body Movement Effects on Behavioral Biometrics for Continuous Smartphone Authentication

L. Nguyen

IEEE Transactions on Biometrics, Behavior, and Identity Science · vol. 10, pp. 1–1

Continuous authentication re-checks who you are at intervals after login, using features that capture how you behave. Its usual cost is latency: behavioural features have to be gathered for 45 to 120 seconds before error rates become acceptable, which means an attacker is already well inside the session before anything is detected.

Equal error rate, and how long it takes to get there

Swipe and phone movement · 60-second window

15.7%

Body motion and posture · 1-second window

6.4%

What we did

3D motion capture data turned into an extensive set of body motion and posture features, aimed at low error rates at 1 to 5 second latencies. Data came from 39 users performing smartphone tasks in a motion capture studio — 41 IR-reflective markers on the body and 3 on the phone, tracked by motion capture cameras, with subjects either sitting or walking a set path.

Findings

  • A 6.4% equal error rate at 1-second latency, and 5.4% at 5 seconds.
  • Swipe and phone-movement features alone reached only 15.7% EER, and needed a 60-second window to do it.
  • Collected from 39 users, with the dataset released publicly.
  • The intent is to capture these features with ordinary smartphone sensors and consumer wearables rather than a motion capture studio.
  • Shifts continuous authentication from reactive to proactive — detection at the start of an attack rather than well into it.

Keywords

Behavioral biometrics
Continuous authentication
Body motion
Posture
Motion capture
Smartphone security

Worked in

Python
NumPy
Pandas
Matplotlib
scikit-learn
Jupyter