Markerless motion capture, from a video on your phone

Markerless motion capture measures how a body moves from ordinary video, without attaching anything to the person. Traditional motion capture places reflective markers on anatomical landmarks and tracks them with a ring of calibrated cameras. Markerless replaces the physical markers with a model that infers where those landmarks are directly from the pixels — so one camera, no suit, no marker placement, and no capture volume to work inside.

A pose-estimation neural network, trained on a large set of images with human body landmarks labelled by hand, predicts the position of those landmarks in every frame of your video. That gives coordinates per frame. Joint angles are then computed geometrically from the coordinates — the knee angle is the angle between the thigh and the shank — rather than being predicted by the network. The AI locates the body; ordinary trigonometry does the measuring.

It depends on your footage, and marker-based capture in a lab is still the reference standard. Markerless measures what a single camera can see, so a joint rotating out of the filmed plane reads short, and occlusion, loose clothing, low frame rates and steep camera angles all cost accuracy. Where it is strong is repeatability and ecological validity: the same clip measured twice gives the same numbers, and you can capture in the gym or on the pitch rather than in a lab, which changes how people move. We have not published a validation study against marker-based capture, so we do not claim lab-grade accuracy.

No. A modern phone is enough. Framing matters more than the camera: square-on to the plane of the movement, whole body in shot, fixed position, decent light, and the highest frame rate you can manage.

If you need a reference-standard 3D measurement for a study judged on its absolute accuracy, use marker-based capture in a lab. If you need to measure movement repeatedly, in the place it actually happens, across more people than a lab session can process, markerless is what makes that possible at all. Plenty of people use both — markerless for volume and for tracking change, marker-based when a number has to be defensible in absolute terms.

Our measurements are 2D, taken in the camera's plane from a single video. True 3D markerless capture needs several synchronised cameras and the reconstruction that goes with them. For a lot of practical work — a knee through a squat, a shoulder through a press, an elbow through a curl — a well-framed 2D measurement answers the question, provided the movement happens in the plane you filmed.

Fourteen joint angles on every frame — neck, both shoulders, elbows, wrists, hips, knees and ankles, plus the trunk's lean off vertical — with the minimum, maximum and range for each across the clip or a section you mark. All 33 tracked body landmarks are available per frame too. Everything exports as JSON or CSV.

They are stored against your team's account so you can come back to them, and they are processed to produce the measurement. They are not used to train any model.