How BikeFit/LAB measures your riding position

1. What happens during a measurement

You film about 12 seconds of pedalling on a trainer, or upload a video, or upload four photos at the 12, 3, 6 and 9 o'clock crank positions. The camera should be at a right angle to the bike, 2 to 4 metres away, lens at saddle height, with the whole rider and bike in frame.

The analysis runs in the browser. The video is not uploaded to our servers; only the resulting measurements are sent to generate your report (see the privacy policy).

The model estimates 33 body landmarks per frame. Because a side view hides one half of the body, the software uses the side facing the camera (the side with the higher landmark visibility) and ignores frames in which the shoulder, hip, knee or ankle is not clearly visible.

2. Which angles are measured, and how

• Knee extension the hip–knee–ankle angle at the bottom of the stroke, read as the 97th percentile over all usable frames.
• Knee angle at the top the same angle at its most bent, the 3rd percentile.
• Hip angle the shoulder–hip–knee angle at its most closed, the 5th percentile.
• Torso angle the shoulder–hip line relative to horizontal, median over the clip.
• Elbow angle shoulder–elbow–wrist, median over the clip.
• Cadence counted from the peaks in the knee-angle signal, one per pedal revolution.
• Also reported: knee position relative to the pedal axle, hip travel on the saddle and ankle range. An optional front-view video adds knee tracking and pelvic tilt.

Percentiles are used instead of a smoothing filter on purpose: the lowest and highest few percent of frames absorb isolated tracking errors, while a smoothing filter would flatten exactly the peaks that matter for saddle height.

3. From degrees to millimetres

Angles alone do not tell you how far to move a saddle. You enter your inseam once; the software estimates your thigh length from it (about 0.54 × inseam), measures the thigh in pixels across the whole clip, and uses that ratio as the scale for millimetre suggestions. An error in the inseam therefore carries through into every millimetre value.

4. Reference ranges used by the software

The report marks a value as on target inside the range below and as close within 4° of it. These are the ranges the software uses today.

Other checks: cadence 70–105 rpm; knee over pedal axle within ±20 mm; hip travel up to 25 mm.

5. Where these ranges come from

Knee angle (saddle height), research-based. A 2022 systematic review found strong evidence that saddle height should be set from the knee angle measured while pedalling (Bini & Priego-Quesada, 2022). An earlier review suggested about 25–30° of knee flexion at the bottom of the stroke (Bini et al., 2011); knee angles measured while pedalling come out larger than static ones, and a study comparing both methods used 30–40° of flexion as the dynamic reference (Ferrer-Roca et al., 2012). In the angle convention above, 30–40° of flexion is about 140–150° of knee extension. The race range sits inside that band; the endurance, comfort and city ranges deliberately allow a slightly lower saddle, which is a practice choice, not a research finding.

Small changes. The same 2022 review found only moderate evidence that saddle-height changes smaller than about 4% of leg length change joint loads more than trivially, and limited evidence on injury risk. Millimetre suggestions below that size are about comfort and consistency, not about proven injury prevention.

Everything else, practice-based. The hip, torso and elbow ranges, and the separate ranges for mountain bike, city bike and time trial, follow common bike-fit practice. They have not been tested in controlled trials, and we present them as starting points, not as norms.

What the evidence does not show. Systematic reviews find that the link between individual bike-fit measurements and knee pain is weak and based on small studies (Johnston et al., 2017; Bini & Flores Bini, 2018). A value outside a range is a reason to look at your position, not a diagnosis.

6. How accurate is it?

We have not published a validation study of BikeFit/LAB itself. The closest independent evidence is a study of 26 cyclists that compared MediaPipe (the model we use) with reflective markers on standard 30 fps side-view video: typical errors were about 3–6° at the knee and hip and 5–9° at the ankle, and MediaPipe showed the knee as more bent than the markers in the part of the crank cycle that includes the bottom of the stroke (Bini et al., 2024), which is where knee extension is read. If that bias applies to your video, knee extension is slightly underestimated.

What that means for your report: treat knee and hip values as estimates with a margin of several degrees, treat the ankle value as indicative only, and use repeated measurements under the same setup to judge whether a change moved you in the right direction.

7. Limits of a 2D camera measurement

• A single camera sees one plane. If the camera is not at a right angle to the bike, or not at saddle height, angles shift. The capture screen checks for a pure side view before recording.
• The hidden side of the body is estimated by the model and is not used for angles.
• Loose clothing, poor light and a busy background make landmarks less stable.
• Millimetre suggestions depend on the inseam you enter.
• Software cannot assess flexibility, injuries, foot anatomy, saddle pressure or how a change feels over a long ride. A professional bike fitter or, for pain, a physician or physiotherapist can.

8. Who writes this, and how

BikeFit/LAB is built by software developers, not by certified bike fitters or clinicians (see About). Guides list the publications they rely on at the end of each article, and every guide shows its last revision date. Statements without a listed source reflect common bike-fit practice, not established research.

General position guidance, not medical advice. See the medical disclaimer.

References

1. Bini R, Priego-Quesada J (2022). Methods to determine saddle height in cycling and implications of changes in saddle height in performance and injury risk: a systematic review. Journal of Sports Sciences, 40(4):386–400. doi:10.1080/02640414.2021.1994727
2. Bini R, Hume PA, Croft JL (2011). Effects of bicycle saddle height on knee injury risk and cycling performance. Sports Medicine, 41(6):463–476. doi:10.2165/11588740-000000000-00000
3. Ferrer-Roca V, Roig A, Galilea P, García-López J (2012). Influence of saddle height on lower limb kinematics in well-trained cyclists: static vs. dynamic evaluation in bike fitting. Journal of Strength and Conditioning Research, 26(11):3025–3029. doi:10.1519/JSC.0b013e318245c09d
4. Bini RR, Nascimento VB, Nibali A (2024). Validity of neural networks in determining lower limb kinematics in stationary cycling. Sport Sciences for Health, 20:127–136. doi:10.1007/s11332-023-01075-7
5. Johnston TE, Baskins TA, Koppel RV, Oliver SA, Stieber DJ, Hoglund LT (2017). The influence of extrinsic factors on knee biomechanics during cycling: a systematic review of the literature. International Journal of Sports Physical Therapy, 12(7):1023–1033. doi:10.26603/ijspt20171023
6. Bini RR, Flores Bini A (2018). Potential factors associated with knee pain in cyclists: a systematic review. Open Access Journal of Sports Medicine, 9:99–106. doi:10.2147/OAJSM.S136653
7. Clarsen B, Krosshaug T, Bahr R (2010). Overuse injuries in professional road cyclists. American Journal of Sports Medicine, 38(12):2494–2501. doi:10.1177/0363546510376816
8. Martin JC, Spirduso WW (2001). Determinants of maximal cycling power: crank length, pedaling rate and pedal speed. European Journal of Applied Physiology, 84(5):413–418. doi:10.1007/s004210100400
9. Brisswalter J, Hausswirth C, Smith D, Vercruyssen F, Vallier JM (2000). Energetically optimal cadence vs. freely-chosen cadence during cycling: effect of exercise duration. International Journal of Sports Medicine, 21(1):60–64. doi:10.1055/s-2000-8857

2026-09-27