How accurate is smartphone gait analysis compared with 3D motion capture?

Why there is no single accuracy figure

A smartphone gait assessment does not produce one number; it produces a set of measures, each derived differently from the same signal. Some depend mainly on detecting the timing of gait events, which inertial sensors do well. Others require estimating distance or separating overlapping phases of the gait cycle, which is considerably harder.

Any honest answer to “how accurate is it?” therefore has to be given measure by measure. A single headline figure either flatters the weak measures or understates the strong ones.

How agreement is expressed

Validation studies commonly report the intraclass correlation coefficient, or ICC, which describes how closely two measurement methods agree. Values run from 0 to 1, and higher is better. Interpretation is conventionally banded — broadly, values in the 0.5 to 0.75 range indicate moderate agreement, 0.75 to 0.9 good agreement, and above 0.9 excellent agreement.

ICC describes agreement with the reference method in the population and conditions tested. It is not a statement of clinical validity, and it does not transfer automatically to a different population, placement or environment.

Published agreement with 3D motion capture: MoveLab 2025

In MoveLab’s 2025 peer-reviewed study, smartphone-derived gait measures were compared against Qualisys marker-based 3D motion capture in 25 healthy adults, with the phone secured in a pouch at the waist, in a controlled laboratory.

The strongest measures

Temporal measures performed best. Cycle time, step time, stance time and cadence all reached good agreement, which is consistent with inertial sensing being well suited to detecting when gait events occur.

• Cycle time 0.894 and step time 0.866 — good agreement.
• Stance time 0.864 and cadence 0.859 — good agreement.
• Gait speed 0.761 and step length 0.745 — around the boundary between moderate and good agreement.

The weaker measures, and why they matter

Double-support measures were the weakest in the study, and this should not be glossed over. Double support is the part of the gait cycle when both feet are on the ground, and it is clinically interesting precisely because it tends to lengthen when people are unsteady or cautious.

• Terminal double support 0.430 — the lowest agreement of any measure tested.
• Double support 0.485 — below the moderate-agreement range.
• Initial double support 0.501 — at the lower boundary of moderate agreement.
• Swing time 0.676 and stride length 0.590 — moderate, and weaker than the leading temporal measures.

How to describe the results

Measures with good agreement against a laboratory reference in healthy adults, under controlled conditions, with a defined phone placement — that is what the evidence supports, and it is a genuinely useful thing to be able to say.

Used the other way round — as a repeatable, objective measure of change in the same person over time — smartphone measurement is comparatively well suited to the job, because each assessment is taken with the same method under the same protocol.

Read the primary source

Every figure above is taken from the published study. The full method, protocol and interpretation are set out on the publication page, and the paper itself is open access.