Automatic archery target scoring: Zin 2013 study

A group from Osaka City University proposes a system that reads the score from video footage of the target. The programme separates from the background whatever changes between one frame and the next, cleans up the image with morphological operations, reconstructs the rings as concentric ellipses and locates the arrow point where the line of the shaft meets the line of its shadow. The score is that of the ellipse containing the point, from 10 to 1.

In the test four archers shot on two different days at 10 metres, each with 6 arrows of different colours, and the system placed every arrow in the right ring: the reported accuracy is 100%. The paper does not say how many arrows were analysed in total, however, and the only example shown is an end of six arrows all in the gold, four in the 10 and two in the 9.

The footage was taken with two cameras, 5 and 10 metres from the target and 1.3 metres high, and frames were extracted from the video at 800 by 640 and 400 by 320 pixels. The authors followed Olympic rules except for the distance, reduced from 70 to 10 metres. The background model is updated frame by frame, and the difference from the two previous frames is used to tell the newly arrived arrow from those already in the boss; thresholds are calculated automatically with Otsu's method, to cope with changes in light. Accuracy is defined as the ratio of arrow points placed in the right area to the total number of arrows on the target.

The test is small and the conditions easy: ten metres, indoors, few arrows, and no information on bow type, target face or the archers' level. A 100% on an unstated number of arrows, with an example in which every arrow is in the gold and well spaced, says little about how the system would cope with arrows close together, overlapping or on a line. The line-cutting rule is mentioned in the introduction, but the method scores the arrow from the position of its point without taking the shaft's thickness into account.

The authors themselves acknowledge that outdoors, with changing weather and light, recognising colours becomes difficult, and that the arrow's movement on impact can shift the estimated point; for both they refer to future work. The comparison with other systems amounts to saying that none had reached 100%, with no tests on the same data.

For archers nothing changes: this is a prototype tested on a handful of arrows at 10 metres, useful mainly for understanding how camera-based scoring systems work and why they are less simple than they look. The most curious detail is the way the point is found, by looking for where the shaft crosses its shadow, which only works if there is a shadow and it is clearly visible. Some of the authors were also working on sensor-based sighting systems for blind and visually impaired archers, which the introduction cites among the archery research the work builds on.