How it works
Two 30-second takes, read on the phone
The phone runs a pose model (MoveNet Lightning) on the video, frame by frame, and keeps twelve points per frame: shoulders, elbows, wrists, hips, knees, ankles. Everything below is measured from those twelve points. The video itself never leaves the device.
Filming
Side view
- Phone on a tripod or shelf at hip height, about 3 metres from the treadmill, square to it.
- Landscape. The whole runner in the picture, head to feet, with some belt in front and behind.
- Runner at a steady, comfortable pace for at least a minute before you start.
Rear view
- Phone on a tripod at hip height, about 2–3 metres straight behind the treadmill.
- Portrait. The whole runner in the picture, head to feet.
- Shorts or tights that show where the hips and knees are help the reading.
The take is read at 30 frames a second. A video with fewer than 20 distinct frames a second is refused — contact would be missed between frames — and under 28 the report says the angles are rougher than usual. A take needs at least 6 steps and a steady rhythm; a longer clip is read for its first30 seconds.
The side view
On a treadmill the body stays in place and the feet swing past it, so each ankle's position relative to the hip goes forward and back once per stride. tiag takes foot contact as the moment the leading ankle is furthest forward, and toe-off as the moment the trailing ankle is furthest back. Both are close to the real events; the foot usually reaches its forward-most point a moment before it lands, so the ankle-ahead distance is, if anything, slightly overstated.
- Cadence: Steps per minute, both feet, from the back-and-forth of the ankles across the whole take. It comes from the whole take at once (how often the movement repeats), not from timing single steps.
- Knee bend at foot contact: How bent the knee is as the foot lands (hip–knee–ankle). 0° is a straight leg.
- Ankle ahead of hip at contact: How far in front of the hip the ankle is as the foot lands, as a share of the runner's leg length. The leg length is the hip-to-ankle distance with the leg at its straightest in the same take, so the zoom and the distance to the treadmill do not matter. With the runner's height entered, it is also shown in centimetres (leg length taken as 49.1% of height, from Drillis & Contini).
- Hip extension at toe-off: How far the thigh is behind the line of the trunk as the foot leaves the belt (shoulder–hip–knee).
From the side one leg hides the other, and pose models often swap left and right from frame to frame. So the side view does not rely on them: the leading ankle is whichever is in front. Side-view numbers are for both legs together; left and right are compared in the rear view.
The rear view
Filmed from behind with the phone's back camera, the runner's left is on the left of the picture, and the hips never swap sides — so tiag decides left and right from the picture, not from the model's labels. The foot lower in the picture is the one on the belt.
- Pelvic drop: in each stance, the peak drop of the opposite hip, measured against the runner's own average hip line through the take — so a phone that is not quite level does not change it. The price: a pelvis that sits tilted the whole time reads as level.
- Knee to midline: in each stance, how close the stance knee comes to the middle of the pelvis, in half hip widths. 1 is straight under its own hip, 0 is on the midline, below 0 is across it.
Every value on the report is the median over all the steps read, with the middle half of the steps beside it, so one odd step does not move it.
Thresholds
Flags mean "worth a look", not that anything is wrong. Each value says where it comes from.
| Flag | Value | Basis | Why |
|---|---|---|---|
| Overstride: ankle ahead of hip | 0.35 leg lengths | Our choice | Ankle more than about a third of a leg length ahead of the hip at contact. Our own cautious choice: plenty of healthy heel-strikers land a quarter to a third of a leg length ahead, and contact is estimated as the ankle's forward-most point, which slightly overstates the distance — so the cut-off sits on the generous side. |
| Overstride: knee bend at contact | 15° | Our choice | Knee bent by 15° or less at contact. Our own choice, chosen so that the flag needs both a long reach and a nearly straight knee before it appears. |
| Pelvic drop | 7° | Our choice | Peak drop of the opposite hip during stance, against the runner's own average hip line. Our own choice, informed by the direction of Bramah 2018; to be calibrated in the pilot. |
| Pelvic drop, left vs right | 3° | Our choice | A difference between sides this big is larger than the take-to-take wobble we expect from a phone. Our own choice. |
| Knee crossover | 0 half hip widths | Our choice | The knee reaching the midline of the pelvis during stance. A geometric definition rather than a clinical one: in the picture, the knee crosses the middle of the body. |
Cadence, knee bend and hip extension are reported without a flag. There is no single right number for them, and the useful comparison is a runner against themselves, before and after.
Sources
- Heiderscheit BC, Chumanov ES, Michalski MP, Wille CM, Ryan MB. Effects of step rate manipulation on joint mechanics during running. Med Sci Sports Exerc. 2011;43(2):296–302. — Raising step rate by 5–10% reduced the load on the hip and knee in that study. tiag reports cadence and does not flag it: there is no single right number, and the useful comparison is a runner against themselves, before and after.
- Bramah C, Preece SJ, Gill N, Herrington L. Is there a pathological gait associated with common soft tissue running injuries? Am J Sports Med. 2018;46(12):3023–3031. — Injured runners in that study landed with the foot further ahead of the body and a straighter knee, and showed more contralateral pelvic drop, than uninjured runners. That is why these are the measures tiag looks at. The cut-offs below are our own, not theirs.
- Souza RB. An evidence-based videotaped running biomechanics analysis. Phys Med Rehabil Clin N Am. 2016;27(1):217–236. — A clinician's guide to reading running from 2D video: which camera views, which moments of the stride (foot contact, midstance, toe-off) and which angles. tiag follows the same views and moments.
- Drillis R, Contini R, Bluestein M. Body segment parameters: a survey of measurement techniques. Artif Limbs. 1964;8:44–66. — Segment lengths as fractions of standing height. Used only to turn a distance measured in leg lengths into an approximate number of centimetres when the runner's height is entered.
What it cannot see
Nothing below the ankle: heel or forefoot strike, the foot turning out, the arch. Nothing in depth: a flat picture shows the body's movement projected onto it, so the angles depend on the camera being square to the runner. tiag gives estimates, not a medical assessment.