Attention Temporal Graph Convolutional Network for Tennis Groundstrokes Phases Classification
Fragment książki (Materiały konferencyjne)
MNiSW
140
konferencja
| Status: | |
| Autorzy: | Skublewska-Paszkowska Maria, Powroźnik Paweł, Łukasik Edyta |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Język: | angielski |
| Strony: | 1 - 8 |
| Web of Science® Times Cited: | 0 |
| Scopus® Cytowania: | 2 |
| Bazy: | Web of Science | Scopus |
| Efekt badań statutowych | NIE |
| Finansowanie: | The research programme titled "Biomechanical parameters of athletes in individual exercises" and based on the analysis of 3D motion data and EMG, realised in the Laboratory of Motion Analysis and Interface Ergonomics was approved by the Commission for Research Ethics, No. 2/2016 dated 8.04.2016. |
| Materiał konferencyjny: | TAK |
| Nazwa konferencji: | IEEE World Congress on Computational Intelligence 2022 ; IEEE International Conference on Fuzzy Systems (FUZZ-IEEE) 2022 |
| Skrócona nazwa konferencji: | IEEE WCCI 2022 ; FUZZ-IEEE 2022 |
| URL serii konferencji: | LINK |
| Termin konferencji: | 18 lipca 2022 do 23 lipca 2022 |
| Miasto konferencji: | Padwa |
| Państwo konferencji: | WŁOCHY |
| Publikacja OA: | NIE |
| Abstrakty: | angielski |
| Action classifi cation makes signifi cant contributions to sport analysis, especially for athletes’ training evaluations. In this study, the identifi cation of two basic tennis groundstrokes were taken into consideration, because these types of movements are performed in every match or training. As the input dataset, the recorded images presenting the registered movements with the optical motion capture system were applied. They show a three dimensional model of the tennis player together with a tennis racket. The Attention Temporal Graph Convolutional Network was used as a classifi cation tool for this sophisticated data. This type of classifi er has an advantage that each graph (joint) in the image corresponds to the human topology. The presented approach is the fi rst application for this type of data, according to the authors’ knowledge. In this study, each forehand and backhand shot from input data was divided into the following phases: The preparation and the point of impact together with the racket swinging. Additionally, no shot was also taken into consideration. Due to the fact that the analysed groundstrokes are very fast and dynamically performed, the strict boundaries of these phases are hard to indicate. Fuzzifi cation of the input data was applied in order to reduce this disadvantage. This approach improved the accuracy with respect to the non-fuzzy input. The obtained accuracy with fuzzifi cation was achieved at the level of 86.9% — 93.82%, while without fuzzy inputs at 74.22% – 81.95%. |
