The study matched 2D video to a lab system. Does that mean my phone is accurate?

Not directly. That study used a fixed camera at 4 m, 1 m high, shooting 240 frames per second with a 1/2000 shutter, on twelve golfers. A phone at 30 or 60 frames per second, propped at an unknown distance, is a different instrument, and no phone was tested.

Why does my shoulder turn read differently on two videos of the same swing?

Shoulder turn is mostly rotation toward and away from the lens, which is the axis a single camera has to infer rather than see. Face-on and down-the-line footage infer it from different information, so they disagree. Compare turn numbers only against your own swings filmed from the same spot.

Should I ignore the club speed number entirely?

Use it as a rough bracket and nothing more. It is inferred from how your body moves, not from tracking the clubhead, so it cannot tell you whether a change added three miles per hour. For that question you need a launch monitor or a radar.

Filming and measuring

What a phone can actually measure in your swing

By the end of this you will know which numbers from a phone-based swing analysis you can act on, which ones carry an error bar, and which ones are guesses wearing a decimal point.

A smartphone mounted on a small tripod on the grass, filming a golfer mid-swing on a driving range. Retro 56, CC0

Angles inside the frame are the honest part

240 fps, 1/2000 shutterCamera in the 2D vs 3D study, twelve golfers

A phone films a three-dimensional motion and saves a flat picture of it. Anything that lies inside that flat picture, like spine angle at address from face-on or shaft angle at halfway back from down the line, is a matter of counting pixels.

Researchers compared three low-cost 2D video packages against a Vicon 3D motion system, twelve golfers, five swing phases. Club angle correlations against Vicon ran from r = 0.920 to r = 0.999, and total swing time showed no significant difference between the 2D and 3D systems. Read the setup before you borrow the confidence. The camera sat 4 m from the golfer, 1 m high, at 240 frames per second with a 1/2000 shutter. No phone was tested.

Toward the camera and away from it, everything is an estimate

Two of threeCoordinates a single lens can actually see

Pose models like MediaPipe output three coordinates for every joint from a single flat image. The third one, depth, is inferred from that image rather than seen, because one lens has no second view to compare against.

That is why hip depth, shoulder turn in degrees, and how far your head moved off the ball can read differently on a face-on video and a down-the-line video of the same swing. Where to stand and how high to hold the phone belongs to the filming guide.

Hear it

Three clicks: takeaway, top, impact. Start at the tour beat and move to whichever one you can actually swing to.

Club and ball speed are inferences with wide ranges

About 4.5 feetClubhead travel between frames at 30 fps, from 93.4 mph

Take the average male amateur driver speed of 93.4 mph. That is about 137 feet per second, so the clubhead covers roughly four and a half feet between two frames at 30 fps, and a bit over two feet at 60 fps. The head is a smear across the frame, and the frame where it meets the ball usually does not exist.

Nothing in that video pins down the speed at contact. Pebble Swing reports club and ball speed as ranges for this reason, derived from body motion and tempo rather than from seeing the clubhead at impact. Treat the range as a bracket. If the number matters, hit balls in front of a launch monitor.

What repeats across clips is your swing

27 golfers, 20 swings eachMarkerless study sample

A single-video markerless study filmed 27 golfers, 20 swings each. Skill did not show up in the forward tilt angle itself. It showed up in how little that angle varied between trials. The better players repeated themselves.

The study did not prescribe a number of clips to film. Three is my own floor, because measurement error in a video is partly random and moves from swing to swing while your habit does not. A fault that shows in all three clips is your swing. A fault that shows in one is probably the light, the camera position, or the frame the software picked.

Questions

The study matched 2D video to a lab system. Does that mean my phone is accurate?: Not directly. That study used a fixed camera at 4 m, 1 m high, shooting 240 frames per second with a 1/2000 shutter, on twelve golfers. A phone at 30 or 60 frames per second, propped at an unknown distance, is a different instrument, and no phone was tested.
Why does my shoulder turn read differently on two videos of the same swing?: Shoulder turn is mostly rotation toward and away from the lens, which is the axis a single camera has to infer rather than see. Face-on and down-the-line footage infer it from different information, so they disagree. Compare turn numbers only against your own swings filmed from the same spot.
Should I ignore the club speed number entirely?: Use it as a rough bracket and nothing more. It is inferred from how your body moves, not from tracking the clubhead, so it cannot tell you whether a change added three miles per hour. For that question you need a launch monitor or a radar.

Sources

Every number on this page comes from one of these.