Classification of Handball Player Ability During Overarm Throwing Motion

Assessment of the quality of sports motion is very important for providing individual training and adapting to current performance. In handball, a good shooting motion is key during a match. In this study, a method is proposed to automatically classify expert and beginner handball shooting motions. This method is illustrated using the public H3DD dataset. This dataset contains the motions of 44 beginners and 18 experts. Each shooting action is represented by a sequence of 3D skeletons. The proposed method uses a dynamic time-warping algorithm, with three angles extracted from each frame. These three angles were selected by expert handball coaches as representatives of the shooting motion dynamics in each frame. This is the first time that handball player ability has been automatically classified during a shooting motion using the aforementioned public dataset. The proposed method achieved an average accuracy of 90.47% for the test set (10% test, 90% training) by using randomly selected balanced samples from the dataset.