Walkfigure takes an ordinary video of someone walking past a phone. It finds the body’s keypoints, measures the hip, knee, ankle and trunk angles and the timing of every step, and shows the walk as a 3D figure you can turn, step through frame by frame, and stand beside a typical walker — with Somi, an assistant that reads the numbers with you. The video is analysed on your own computer. It is not uploaded.
Research software — beta Free to download
Walkfigure is research software in beta. It is not a medical device, it does not diagnose, and it does not recommend treatment. Its numbers are estimates to check against the video, not measurements to rely on.
No markers, no suit, no second camera. One phone, one walk across the picture, and the computer you already have.
Stand side-on to the path and keep the whole body in view. The person walks across the picture. An ordinary phone video is enough. Where a video cannot be measured — nobody in the picture, several people at once, a phone turned on its side, too few strides — it says which of those it found, and what to do about it.
It finds and tracks the person, places the keypoints, detects the heel strikes and toe-offs, and works out the angles and the timing. It first shows the height and build it has estimated, beside a frame with the face blurred, and asks whether they look right.
The walk opens as a 3D figure. Turn it, slow it to a tenth of the speed, step through a frame at a time, jump between heel strike and toe-off, stand a typical walker beside it, ask Somi what a number means, and export a report or the numbers behind it.
Measurements first, then ways of looking at them. Where a part of the program is a sketch rather than a measurement, it says so, and so does this page.
The angles are the program’s own keypoint geometry, in the side view only. Distances and speeds rest on the height you give it; angles and timing do not.
At each hip, knee and ankle lies a wedge in the limb’s colour. It runs from a dashed line, the line the angle is measured from, round to the limb itself, so a knee bent 45° shows 45° of wedge and the number can be checked by eye. The trunk’s lean is drawn the same way, from straight up.
One button stands a typical walker beside the figure. It is the person’s own size, segment for segment, and its heel strikes fall on the person’s own, so the two are always at the same point of their stride and can be compared joint for joint.
Its hips, knees and ankles follow the mean curves of a published reference group (Lencioni et al. 2019, CC BY 4.0). It is a construction from the averages of a small group. It is not a person and not a goal: nobody walks exactly the mean.
Any joint or segment angle from heel strike to heel strike: each stride, their mean ± SD, or both, with toe-off marked. Typical walking sits behind the hip, knee and ankle as a grey band, and the panel says under the chart what the band is and where it came from.
The reference was measured with skin markers in gait laboratories, so offsets of several degrees from keypoint angles are expected, most at the ankle.
The small grey box at the bottom left of the 3D view is Somi, Walkfigure’s assistant. Ask it to measure an angle, explain a number, or set up the scene — place figures, change how they look, move the camera, run the physics — in plain words, and it does it on this walk and this scene.
/ lists its commands, ? the keyboard shortcuts./key and have an allowance to themselves.Play lets you walk the figure about yourself, with W A S D or an Xbox controller, on its own recorded stride. It can jump, wave and dance, and nothing it does is kept.
Kids’ mode is for a gait laboratory. A child sees the figure that is their own walk, brightly coloured under a sky on grass, and plays with it as they would a video game. Everything clinical is put away, nothing they do is saved, and leaving takes a grown-up’s held press.
The recording plays beside the 3D view, held to the same frame, with the skeleton drawn over it. Faces are blurred unless you untick it, and a clip recorded from the panel is blurred too. Measure a length or an angle on the picture.
Push a figure, throw a ball at it, or put a box or a slippery patch in its path. It staggers, catches itself with a step or goes down, and a fall report says what happened and when. It is a physics sketch, not a model of the person — for exploring and explaining, never for estimating anyone’s risk of falling.
A report as PDF or HTML with the caveats on its first page; CSV of every frame, of the strides and of the summary; the motion as BVH for animation tools; a clip or a snapshot of the view. A REST service on the same machine, with SDKs for Python and JavaScript, runs the same analysis for other programs.
People who look at walking and have to explain what they see. It is a tool for looking and explaining, not for deciding what is wrong with someone.
Something to show the person who has just been measured: their own walk as a figure they can turn round, beside a typical walker. Kids’ mode is for the child who would rather play than look at curves.
A way to look at a walk with the person in the room. Slow it down, turn it, point at the knee, and keep each walk as a project to look back on.
Frame-by-frame CSV, per-stride tables, BVH motion and a local API. For teaching, a figure that students can step through, measure, and push over.
A video of someone walking is a video of a person. The analysis happens where the video already is.
The analysis runs on your computer. There is no cloud service behind it, the engine needs no internet connection, and in a checked run it made none.
Each project is a folder on your own machine. A video can be taken away and the analysis kept, and a deleted project goes to the Recycle Bin.
Faces are blurred in the video panel, in a recorded clip and on each project’s card. Walkfigure never works out who is in a video: a project carries a person’s name only if you type one.
The one exception is Somi, and it is stated plainly above: what you ask it, and the numbers it looks up to answer, go to the model provider. Leave Somi alone and nothing leaves the machine.
The program says what it measures and what it fills in. So should this page. Walkfigure is research software in beta: it is not a medical device, it does not diagnose or recommend treatment, and its numbers are estimates to check rather than measurements to rely on.
One setup file, installed for one Windows user. It downloads nothing while it installs, and it needs no administrator.
Free no account, no licence key
One setup file for Windows. Everything on this page is in it: the measurements, the 3D view, the comparison, Somi, Play and kids’ mode. There is nothing to buy, nothing to sign in to, and no part of it that is held back.
Using it in a clinic, or want it on a walk of your own first? Write to contact@mgam.bio.
Version 1.1.69, 407 MB. Windows 10 or 11, 64-bit.
Walkfigure is not code-signed, so Windows will warn you the first time: a blue “Windows protected your PC” box. That warning means Windows does not recognise the publisher — not that anything is wrong with the file. To run it anyway, choose More info, then Run anyway.
If you would rather check the file first, its SHA-256 is below. In PowerShell, Get-FileHash WalkfigureSetup.exe should print the same string.
706cb44e99efdbab475ea9348b468f13093dfada83a5194f59a391da65280e52Walkfigure is archived on Zenodo and has a DOI of its own. If it did any of the work behind something you publish, cite it — and cite the work it stands on.
10.5281/zenodo.22837820 always resolves to the newest version, and is the one to cite unless you need to name the exact version you ran — every release has a DOI of its own on the Zenodo record.To see it on a walk of your own, or to ask anything about it, write to us and say what you would use it for.