Range of motion is one of the oldest measurements in physiotherapy and still one of the most useful. In clinic, a therapist measures it with a goniometer, a hinged protractor held against the joint. At home, a camera-guided app estimates the same angle from video. Patients and therapists both reasonably ask how far to trust that number. The research gives a fairly clear answer: good enough to track change over time under consistent conditions, not good enough to replace a clinical measurement, and noticeably worse for some joints and camera angles than others.
How a camera turns video into an angle
Pose estimation models find landmarks in each video frame (hip, knee, ankle, shoulder, elbow, wrist) and connect them into a stick figure. The angle at a joint is then simple geometry between three points. Everything depends on where the model places those points. In a biomechanics lab, reflective markers are stuck to the skin and tracked by a dozen or more cameras. A phone has one camera and no markers, so the model has to infer where the joint centre is from how you look, and it was usually trained on photos of healthy people in everyday settings, not on patients.
What the validation studies found
Researchers at the University of Bath compared a single 2D camera against a 15-camera marker-based system while 15 healthy adults walked. On the side of the body facing the camera, hip and knee angles viewed side-on came close to the error range already reported for marker-based systems themselves. On the far side of the body, partly hidden from the camera, the spread of error for hip and knee roughly doubled. The ankle was the weak point: limits of agreement of about 12 degrees either way, which the authors judged too wide for practical use.
Simpler photo-based measurement has been tested against the goniometer too. In a 2018 study of 64 knees, angles measured from smartphone photos differed from goniometer readings by an average of 5 degrees. Photos taken by participants themselves agreed closely with photos taken by a surgeon, with an intraclass correlation above 0.9. The authors called the method valid and reliable, and the self-photo result matters for home use: people can capture their own joint consistently.
A 2022 scoping review from the same Bath group summarised the field. Timing and step-length measures from markerless systems are comparable to marker-based ones. Joint centres and joint angles, in their words, are not yet accurate enough for clinical use in general. The review also makes an awkward point: the marker-based gold standard carries its own error, so nobody knows the true accuracy of either method without X-ray based imaging.
The goniometer is not a fixed reference either
When one therapist measures the same knee twice, agreement is usually excellent. When two different therapists measure it, agreement drops, and a difference of several degrees between two clinicians is common. Knee extension near zero is harder to measure than flexion in the middle of the range. None of this makes the goniometer bad. It means a few degrees of disagreement between any two methods is normal, and the question is whether a method is consistent with itself.
What this means for home tracking
- Trend, not snapshot. A single camera reading sits inside an error band of a few degrees at best. The same movement, filmed the same way, every few days is informative because the error stays roughly constant and the change is real.
- Camera side matters. The joint nearest the camera is measured best. For a knee bend, stand side-on with the working leg closest to the camera.
- Keep the setup fixed. Same spot, same distance, same lighting, similar clothing. Baggy or dark clothing hides joints and moves the landmarks.
- Expect less from ankles and from twisting movements. These are where single-camera estimates fall furthest behind.
- Compare like with like. A home camera number and a clinic goniometer number are two different instruments. Compare home with home, and clinic with clinic.
How KineTrue uses the number
Each KineTrue session logs range of motion per exercise alongside reps or hold time, a movement quality signal, and your pain rating. Your therapist sees the sequence across weeks, not one reading, so a genuine change in range stands out even when any single reading is a few degrees off. It supplements what they measure in clinic; it does not replace it. For more on what else the camera can and cannot tell you, read what a form quality score measures.
If your range of motion drops suddenly, or a joint becomes hot, swollen, or locks, contact your therapist or doctor rather than re-filming the exercise. A camera estimate is a tracking tool, not an assessment.