Validity of OpenPose Key Point Recognition and Performance Analysis in Taekwondo

Performance analysis of sports using computer vision technology has been popular in recent times due to its cost efficiency, accessibility, and applicability. Despite its popularity, technological involvement is limited in its performance analysis in Taekwondo. Therefore, this study aimed to validate the usage of computer vision technology in the performance analysis of Taekwondo. In total, 5 Taekwondo players including 2 world-class and 3 master-level players were recruited for this study. They were asked to perform Taeguk Il-Chang while being captured by 12 VICON and 4 Contemplas cameras. The Contemplas cameras were calibrated using a 1m-squared calibration cage. The Contemplas recordings were processed by OpenPose to extract a 2-dimensional human skeleton which was triangulated to obtain 3-dimensional data. In total, 8 joint angles were calculated from each system and compared using the mean absolute error. The high kick during the performance was analyzed using the computed joint angles. The upper-body joint angles were more erroneous than the lower-body joint angles. The minimum hip angles and high kick completion time were lower and faster in the world-class players than the master-class players. In conclusion, the errors may be by the occlusion. The lower minimum hip angle and faster kick completion in world-class players than master-class players may mean that world-class players are more capable of kicking higher and faster. This fact can be applied to the training design for trainers and players. Despite the conclusion, the small sample size and comparability between OpenPose and a VICON system were considerable limitations.