Camera scoring of archery targets: Phang 2023 study

This describes a prototype for scoring automatically with a camera pointed at the target. The system works in two stages. First, on an image of the face with no arrows, it reconstructs the boundaries of the scoring rings; then, comparing each frame with the previous one, it finds the newly arrived arrow and assigns it to the ring it intersects. The authors deliberately avoid the automatic circle detection used by other systems, which on a worn face filmed at an angle tends to misplace the rings, and instead reconstruct each ring by following its outline.

The paper gives no numbers: no percentage of arrows scored correctly, no count of arrows tested. It shows one session, described as one of the hardest, with a heavily used face and arrows hiding one another, in which the images show correct scores, and one failure: an arrow almost hidden by the others was placed at a different point on the target from where it actually was, although in the same 9 ring.

In the first stage an operator marks by hand the four corners of the face and the radius of the centre on the initial image; from these points the system corrects the perspective, finds the ring edges, cleans them up, smooths their outline by interpolation and takes their convex hull. The camera has to stay still for the whole session. In the second stage each pair of frames is aligned, converted to greyscale and blurred; the difference between the two frames brings out the new arrow, which a line search locates and boxes. The system assumes that only one arrow arrives between one frame and the next.

All the information needed to judge it is missing: the type of camera, the filming distance and angle, the lighting, the number of sessions and arrows, the error rate. It is not known how it handles arrows touching the line between two rings, which under the rules take the higher score. The authors themselves identify arrows hidden by others as the main limitation, and suggest filming the target from the side or from several viewpoints.

For anyone who shoots the study changes nothing: it is about how points are counted, not about shooting. Anyone thinking of using a camera to log their arrows automatically in training will find an idea of how it could be done, not a tested tool, since the paper does not say how often the system makes mistakes.