High-speed video in archery research

What a fast camera measures

A high-speed camera does not measure forces: it records where things are at fixed intervals, and times, angles and speeds are worked out by counting. Times come from the number of frames, displacements from pixels, with a known scale inside the frame. In the study that serves as the reference for the recurve, sixteen national-team archers filmed at 70 m during the London 2011 test event, the camera sat in front of and above the arrow's path at 4,000 frames per second, one image every 0.25 ms at 800 by 800 pixels, and the scale was the diameter of the shaft. The authors estimate errors of 1 to 2 mm on displacements and of half a frame to two frames on the times, and that error, more than the frame rate, sets the limits of what footage can show.

A 1937 model of the bow was built to replace filming at one to three thousand frames per second, which the author had already had made and found slow and costly. A Polish study from 1990 rests on 16 mm film shot at between 1,000 and 2,500 frames per second of a national-team archer and of a shooting machine. The story of that footage, year by year, is in the line on the archer's paradox; this page looks at how it was obtained.

How many frames per second are needed

In the sixteen archers of 2012 the string stays on the fingers for 1.9 ms, the nock reaches its point closest to the body at 5.4, the shaft is straight again at 8.6 and leaves the string at 14.6: this is the claim that about fifteen thousandths of a second pass between the start of the release and the arrow leaving the string. At 4,000 frames per second there are about 58 images between release and separation. At 1,000, the rate used in 2007 on the archer then ranked second in the world, about fifteen remain; at 500, the rate of the infrared cameras that followed the fingers of an Austrian archer in 2009, there is one image every 2 ms, longer than the string takes to leave the fingers.

At 100 frames per second the whole release fits into three images, and the authors of an Italian study from 1994 say so precisely to explain why they leave it out. The same problem weighs on the entry on release duration, which is contested: the three hundredths of a second from 1987 come from an accelerometer and from switches on the clicker and string, the 14.6 ms from 2012 (14.7 in the text of the same paper) come from images, and the two measurements may not start from the same instant.

Speeds are obtained by differentiating position from one frame to the next, and every derivative magnifies the error. In the Austrian archer the middle joint of the middle finger opened at 6,923 ± 1,063 degrees per second in the arrows in the X, with no significant difference from those scoring eight or less, and the authors warn that the derivatives may be imprecise: the entry on finger speed and score remains a finding in a single archer. The same footage supports the claim about which joints open at the release, which relies on angles rather than speeds.

The camera's threshold and its point of view

When footage does not show a movement, the finding holds only down to its resolution. In the sixteen archers of 2012 the bow moved no more than one pixel, 1 mm at most, and rotated less than 0.1 degrees while the arrow was on the string: the entry on the bow staying still while the arrow is on the string says that much, not that the bow is motionless. In 2007, at 1,000 frames per second, the champion's string hand moved back by 1.24 mm while his bow stayed at 0.00.

The viewpoint matters too, because a camera sees one plane. From above, the arrow shows its lateral flex. From the side, in the Japanese software from 2022 that reads the images automatically, at 3,000 frames per second and between 0.69 and 1.1 mm per pixel, an X10 shot by a robot looked up to 2 cm shorter than at rest, because it was flexing and could tilt in the horizontal plane, out of the filmed one: from this the authors derive a flex frequency of about 75 Hz, and they note that seeing the flight in three dimensions needs stereo filming, with a camera added above.

What the footage has shown

The camera took the claim on the archer's paradox to "corroborated", with the nock pushed on average 13.1 mm towards the body and contact with the pressure button over by 8 ms, and it moved attention away from the tail. In the archer whose full sequence is published, the minimum clearance between shaft and button comes at 15.5 ms, about 150 mm from the nock, once the arrow has already left the string, and the nock passes the button only at 18 ms: this is the entry on where the arrow passes closest to the riser. Across the sixteen, tail clearance ranged from about 2 to 25 mm, and only four released the arrow on the bow's median plane.

The same footage explains why a machine can stand in for the archer in these tests: if the bow does not move before the arrow leaves, what counts is how the string leaves the fingers. With a camera directly above the arrow's path, a Japanese laboratory tracked the rear of the arrow frame by frame, fitted a mean line and took the deviation from it as flex: 1.22 mm with a mechanical release, 6.86 with two archers, 14.8 with a finger mechanism. A 2015 robot, filmed the same way, showed that flex does not grow in proportion to finger speed.

Not everything about the arrow comes from video. That the arrow flexes less on a compound rests on timed-flash photographs; that the compound's rotation cannot be cancelled, on accelerometers and a model; the oscillation in flight comes from paper tests and simulations, with cameras giving only the attitude at launch. In flight, cameras measure something else: two stations a known distance apart, one 13 m from the launcher and the other 45 m further on, give the slowing of the arrow and from it the air resistance, on the assumption that the arrow flies without lift. In the 2012 study the angle of attack was estimated from the film to within about one degree, while drag changes regime above six tenths of a degree.

What to check

• how many frames fall within the event being measured, how many millimetres a pixel is worth and what error is stated;
• where the camera looks from, and whether the movement that matters lies in the filmed plane;
• how the zero instant is defined, and whether two studies take it the same way;
• whether speeds are derived by differentiation, and with what filter;
• who is shooting: one archer, sixteen archers, a robot or a compressed-air launcher, which removes the flex;
• what is measured and what is judged by eye: in 2012 finger order, bow torque and face contact were visual judgements, and no footage was linked to where the arrow landed.

How design and sample are weighed is explained in How we read a study, whole-body footage on the page on motion capture, and the arrow from the tuner's side in the articles on the archer's paradox and spine and on the release.

@misc{campagna2026_enresearchmethodshighspeedvideo,
  author = {Campagna, Matteo},
  title = {High-speed video in archery research},
  year = {2026},
  howpublished = {Research, matteocampagna.com},
  url = {https://matteocampagna.com/en/research/methods/high-speed-video},
  note = {accessed 2026-10-16}
}