Speed climbing will appear for the first time at the Olympic Games in Tokyo 2021. It is an exciting discipline suitable for computer-aided processing by employing computer vision and motion understanding methods. This thesis proposes a labeled dataset of speed climbing performances and demonstrates baseline evaluation use cases (e.g.,search, clustering). The proposed dataset contains 362 speed climbing samples (38 minutes in total) of 55 top-performing athletes. The dataset comprises 2D skeleton motion sequences extracted from speed climbing videos. The videos used for skeleton extraction were obtained from the videos of official events (e.g., world championships) that are publicly available on YouTube. The second contribution of this work is the evaluation of k-nearest neighbors search and climbing style analysis. We are able to achieve the retrieval of the same climber's run (1-NN search) with 96% accuracy using the suitable combination of dynamic time warping distance and data normalization.
© Copyright 2021 Veröffentlicht von Masaryk University. Alle Rechte vorbehalten.