Shot consistency in archers by sensor: Quan 2017

Four Korean girls from a middle school, aged 14 to 16, each shot 36 arrows at 30 metres with an inertial sensor on the forearm of the bow arm. The authors compared the acceleration signal of each shot with that of a reference shot, chosen from those that scored 10, using dynamic time warping, an algorithm that measures how alike two curves are even when one runs slower or faster than the other. The smaller the distance, the more the shot resembles the reference: this is the measure of consistency the study proposes.

The two girls with the higher scores had the smaller distances, on average 13.87 (standard deviation 0.9) and 16.72 (1.3); the two with lower scores had larger distances, 23.53 (2.2) and, for the fourth, 23.30 (2.3). The scores the authors refer to are those of the previous season and the current one, out of 60: 59 and 59.5 for each of the first two, 55 and 53.5 and then 54 and 53.7 for the others. Out of a possible 35 shots per archer, leaving out the reference, the algorithm found 35, 35, 29 and 31 in the recording on its own.

The sensor was part of the Noraxon Myomotion system, sampling at 100 Hz, and only the acceleration along the drawing axis was used. The girls shot with their own bows and arrows on the school's outdoor track, in 6 ends of 6 arrows with a short rest between ends; each end took between 110 and 140 seconds. The section analysed covered part of the set-up, the draw, aiming and holding, up to a few tens of milliseconds after the release, and the reference was 6.3 seconds long.

A variant of the algorithm, called aDTW, searched the continuous recording of each end for the six sections most similar to the reference, using a threshold set by hand for each archer; the normalised distance from the reference was then calculated for each section found. The relation between distance and score is shown as a histogram for each archer, with no statistical test. The girls' level is what the authors call elementary, the second step on a four-step scale that runs up to world class.

Four people are too few to speak of a relationship between consistency and score: with two stronger and two weaker archers, the match between distance and score is a description, not a finding. Moreover, the score does not come from these 36 arrows but from the season's results, and no shot is linked to where it landed. The paper does not say whether the girls shot recurve or compound, and gives no draw weights.

The measure has weak points too. The reference is a single 10 chosen by the researchers, so the distance measures similarity to that shot rather than consistency across all shots; the threshold for finding shots is set by hand for each archer; and only one axis of one sensor, on the bow forearm, is used. The shots the algorithm did not find, six for one archer and four for another, are very likely the ones least like the reference, and leaving them out makes the mean distance smaller for precisely the two archers with lower scores.

The idea is of interest to anyone working with sensors: a small accelerometer on the bow forearm, a continuous recording, and an algorithm that finds the shots by itself and says how closely each one resembles a good shot. It is a way of putting a number on the repeatability of the movement, which a coach usually judges by eye.

The study, however, only shows that the method works technically on four archers, not that the measured distance predicts score, nor that working to reduce it improves shooting. The authors' closing remarks about using the method to spot talent or prevent injury have no data behind them in this paper.