Walk across the room with your phone in your pocket or hand. The accelerometer
measures your step cadence, stride length, vertical bounce, and lateral sway.
Together these form a gait signature — a biometric fingerprint that stays stable
across sessions, even as you change shoes, bags, or how you're carrying the phone.
You can be identified from 100 metres of walking with 95% accuracy — no camera needed.
Academic gait recognition using phone accelerometers achieves better than fingerprint recognition
at identifying individuals across a crowd. Chinese street cameras now routinely use gait
recognition as a backup when faces are covered. In 2018, police in Beijing identified a fugitive
who was deliberately obscuring his face, purely from the way he walked.
0
steps detected
put the phone in your pocket and walk
—steps/min
—step Hz
—regularity %
your gait signature (builds as you walk)
step cadence—
stride regularity—
vertical bounce—
lateral sway—
pace type—
fingerprint hash— (need 20+ steps)
▶ what makes a gait biometric unique?
Your walk is shaped by leg length, hip width, muscle tone, posture habits, injury history,
and shoe type. The accelerometer captures this as a time series of acceleration peaks.
From that we extract:
Cadence — steps per minute. Normal walking: 90–130 spm.
Stride regularity — how consistent each step is. Injury shows here first.
Vertical bounce — how much your body moves up and down per step.
Lateral sway — side-to-side motion. Wider gait = more sway.
These features are stable across time and robust to changes in clothes, bags, and phone position.
Real surveillance gait recognition works at 50+ metres using video at low frame rate —
the shape of the curve, not the absolute values, is what gets matched.