Good data starts in the bay.
The most important decisions about your swing data are made before you ever hit a ball — in how the room is lit, what colour the walls are, and how fast the cameras can see. Get the bay right and every measurement downstream gets easier; get it wrong and no algorithm can fully recover. Here's how to set a studio up to be measured, drawn straight from the motion-capture and biomechanics literature.
This is a setup guide, not a hardware spec. The targets below — frame rates, shutter speeds, wall colours — are indicative benchmarks from published motion-capture and golf-biomechanics research: a sensible starting point for any bay, not guarantees for a given rig. The golden rule from that same literature never changes — measure your own setup, and validate it. See how we earn the numbers →
Two angles, one swing
The established minimum for golf is two square-on views. Nothing exotic — but where you place them decides what you can and can't measure.
- 1
Face-on
Set square to the target line, looking straight at the golfer. This is the view that reads sway, lateral weight shift and spine tilt — the side-to-side story of the swing.
- 2
Down-the-line
Set along the target line, behind the hands. This is the view for swing plane and shaft angle — the geometry a coach draws lines on.
- +
More views help in 3D
For markerless 3D reconstruction, extra calibrated cameras cut occlusion and depth ambiguity. Two calibrated cameras is the low-cost baseline; add more where a limb keeps hiding the club.
Illustrative — two orthogonal views, placed to the target line.
Fast enough to freeze it
This is the decision most often gotten wrong, and it follows from a law of physics: to capture a movement cleanly you must sample faster than twice its quickest meaningful part. The club is roughly ten times faster than the body.
- ✓
Body: 240 fps
Pelvis, torso and arms carry only ~10–15 Hz of real content, and the golf literature clusters at 240 fps — comfortable headroom for clean speed and acceleration.
- !
Club: 240 fps is the floor
At 100 mph the clubhead covers ~150 ft/s. At 30 fps it jumps about 5 ft between frames — useless near impact. At 240 fps that drops to ~7.5 inches. Run the club as fast as the hardware sustains.
- 3
Shutter, not just frame rate
A fast frame rate with a slow shutter still smears the clubhead. Set exposure to ≤ 1/1000 s (faster near impact) — the single biggest lever on clubhead precision indoors.
- ✓
Impact is its own problem
Ball contact lasts under half a millisecond; true impact mechanics need hundreds-to-1000+ fps. If you only need impact timing, an audio trigger plus a 240 fps camera is a solid compromise.
Illustrative — the same swing, sampled slow versus fast.
Global shutter, never rolling
One camera-spec decision corrupts golf geometry more quietly than any other: how the sensor reads its pixels out. It has nothing to do with frame rate, and it's easy to get wrong.
- ✗
Rolling shutter bends the truth
A rolling-shutter sensor scans line by line, so a fast shaft is skewed as it's read — a straight shaft photographs as a curve, corrupting the exact swing-plane and shaft-angle you're trying to measure.
- ✓
Global shutter keeps it straight
A global-shutter sensor exposes the whole frame at one instant, so fast-moving geometry stays true. For golf this isn't a nice-to-have — it's a requirement.
Quick check. Wave a straight rod quickly across the frame. If it looks bent or leaning in the recording, the sensor is rolling — and your shaft angles will be too.
Illustrative — rolling shutter skews a straight shaft into a curve.
Bright, steady, flicker-free
Fast shutters are hungry for light, and indoor bays are almost always lighting-limited. The fix isn't simply "more" — it's the right kind.
- ✓
Bright and diffuse
Fast shutters need a lot of light. Make it soft and even — harsh single-source shadows confuse pose estimation and hide the club against the body.
- !
Kill the flicker
Mains-frequency flicker beats against the frame rate and pulses brightness through your footage. Use DC or high-frequency LED / high-frequency-ballast sources, not cheap mains lighting.
- ●
Consistency steadies the AI
Markerless pose estimators degrade badly in shadow and low contrast. Steady, repeatable light makes every measurement that rides on the skeleton more reliable, session to session.
Illustrative — soft even light, and a flat, flicker-free output.
Grey walls, matte finish
No golf-biomechanics paper prescribes a wall colour, but studio practice and the vision literature point the same way — and golf adds one twist that overrides the usual studio default.
- ✓
Matte is non-negotiable
Gloss and satin throw specular highlights, and a bright spot on a shiny wall reads to a detector exactly like a ball, a marker or a clubhead. Matte paint and blackout curtains are the mocap-studio standard.
- !
Go grey, not white
The usual white-wall default is actively wrong for golf, because the ball is white — it blends into a light background and colour segmentation fails. A mid-to-dark grey (or matte black) behind the hitting zone restores ball contrast and soaks up stray light.
- ●
Mid-grey for the swing volume
The shaft is the awkward target — chrome is bright, graphite is dark — so mid-grey gives usable contrast against both, plus against skin and clothing. Pure black maxes out ball contrast but can swallow a dark shaft.
- ✗
Avoid saturated colour
Red, pink, purple and green walls bounce coloured light onto the golfer and wreck colour-based segmentation — unless you're deliberately running a green/blue chroma-key, which is a separate mode, not the default.
Illustrative — ball contrast across wall tones; mid-grey wins.
Keep the scene quiet
Beyond its colour, the background has to behave. Many vision methods assume the scene behind the golfer simply never changes.
- ✓
Uniform and unchanging
Background-subtraction assumes a static, uncluttered scene — anything that moves or shifts between sessions breaks it. Keep the wall plain and the frame the same, day to day.
- !
Clutter looks like clubs
Alignment sticks, trouser creases, turf seams and shadows all look line-like, and tempt a tracker into "finding" a club that isn't there. Clear the camera's field of view of anything you don't need in it.
- ●
The infrared caveat
If you add active IR light or IR markers, a wall's near-IR reflectance doesn't follow its visible colour — a wall that looks dark to the eye can be bright to the sensor. Verify the surround through the actual camera, not by looking at it.
Illustrative — every line-like distractor is a trap for the tracker.
The dark-room equalizer
The room is your choice — nobody should repaint a bay or bin a black-walled sim room to use a camera. But where the room can't give clean light and a quiet backdrop, a few grams of retroreflective tape on the shaft close the gap, and it's the single highest-value optional upgrade a studio can make.
- ✓
What it buys
Retro tape lit by a small light at the lens throws that light straight back and saturates as bright blobs whatever the background — a banded shaft is trackable where a plain one is invisible. In a dark home-sim room it lifts club tracking from the hard case to instrument grade.
- !
The light must sit at the lens
Retro tape returns light to its source within a ~1–2° cone, so a small camera-mounted LED ring does what a ceiling floodlight can't. Visible-band only — never IR, which conflicts with launch monitors.
- ●
Optional, and honest
Vision-only tracking stays the default; taping is an opt-in high-accuracy mode. Every taped swing also doubles as ground truth for proving the untaped tracker, so both get better together.
Marking up a club. The tape is the easy part; the measurement is what makes it worth doing.
1 · The tape. Silver glass-bead retroreflective tape, about 25 mm wide, thin and conformable — not stiff “diamond-grade”, and not plain chrome (both are the wrong kind). A few grams; no effect on swing feel.
2 · The pattern (“2–1–3”, grip → head). Two bands just below the grip, one mid-shaft, three above the hosel — 25 mm bands and 25 mm gaps. The lopsided two-versus-three count tells the software which way round the club is; the lone mid band is occlusion insurance.
3 · Measure it. With the tape on, measure from the butt of the grip to the centre of every band, to the millimetre, and keep that record per club. Known spacings are what let each frame read shaft angle and foreshortening directly — without the record, the tape is only “easier to see.”
4 · Light it. A small LED ring at the camera lens, visible-band (never IR). Set exposure short enough that the bright bands don't clip.
Illustrative — the “2–1–3” pattern (25 mm bands and gaps), measured from the butt.
Strap the IMUs to bone
The rotational quantities golf cares about come from small inertial sensors — and they validate surprisingly well (upper-torso and pelvic rotation agree with optical at ICC 0.91–1.00) if you mount and calibrate them right.
- ✓
Mount over bone, strap firmly
Site sensors over bony landmarks — sternum or thoracic spine for the upper torso, sacrum for the pelvis — with firm strapping to cut the soft-tissue wobble that blurs the reading.
- 1
Calibrate every session
Run a defined static or functional pose each session to fix each segment's frame before the first swing — a second of stillness that anchors everything after it.
- !
Distrust the magnetometer indoors
Studios are magnetically dirty — steel structure, screens, motors. Prefer gyro-dominant yaw with periodic correction, or a magnetometer-free fusion, and check heading stays stable across a session.
- ✓
Fuse fast, don't integrate
Use an orientation filter, not raw gyro integration, which drifts. Run the sensors fast (commonly 100–200+ Hz) so peak-angular-velocity timing — the substance of the kinematic sequence — is well resolved.
Illustrative — the four-sensor proximal-to-distal set.
Line up every clock
A hybrid rig is only as good as its timing and its calibration, and golf's sub-millisecond impact makes both unforgiving. Sub-frame skew between a camera and a sensor scrambles any claim about the sequence.
- ✓
Calibrate the volume every session
Recalibrate the capture volume each session and log the residual. Optical golf setups report sub-millimetre to ~1 mm marker residual; markerless uses a per-session checkerboard with reprojection error logged. Treat anything much worse as a setup fault.
- 1
Share one clock where you can
Use a hardware trigger or common clock — genlock or a shared sync pulse across cameras and sensors — so every stream is stamped on the same timeline.
- 2
Or use impact as a fiducial
No hardware sync? Use one physical event every trial. The impact transient shows up in audio, force and video at once and gives a single sharp mark to align them on.
- !
Measure the skew, don't assume it
Log each stream's timestamps and quantify the residual skew. Cross-modal latency is a first-class error source — a mistimed force trace can invent or hide a kinetic sequence that isn't there.
Illustrative — three rates, aligned on one shared event.
The bay, on one page
Everything above, distilled to a pre-session run-down you can check off before the first swing.
- ✓Two square-on views — face-on plus down-the-line (more cameras for 3D markerless).
- ✓Global-shutter sensors — never rolling shutter.
- ✓Shutter ≤ 1/1000 s; frame rate ≥ 240 fps for body and club.
- ✓Bright, flicker-free, diffuse lighting matched to the fast shutter.
- ✓Matte surfaces throughout — no gloss or satin.
- ✓Mid-to-dark grey (or matte black) walls behind the hitting zone — never white.
- ✓No saturated wall colours unless deliberately chroma-keying.
- ✓Background uniform, unchanging and clutter-free within the camera's view.
- ✓Room chosen consciously — bright light and a quiet backdrop where you can; a dark sim room is fine if you tape the club.
- ✓Dark or busy room? Retroreflective shaft bands (2–1–3, measured from the butt) plus a camera-mounted LED ring — visible-band, never IR.
- ✓IR reflectance verified through the sensor if using active IR.
- ✓Capture volume calibrated every session; residual logged.
- ✓IMUs mounted over bone, calibrated each session, with magnetometer-robust yaw.
- ✓Sensors sampled fast with a proper orientation filter, not raw integration.
- ✓Hardware sync or a per-trial impact fiducial across all streams; skew measured.
- ✓Every figure validated against a reference — the numbers above are starting points, not guarantees.
A room built to be believed.
None of this is exotic — it's ordinary kit, placed and lit with intent. Get the bay right and the hard problems downstream get easier. These targets are drawn from the published literature and offered as a sensible starting point, not a promise; the last word from that same literature holds for every rig — measure your own, and validate it.