Recognition of Tennis Shots Using Convolutional Neural Networks Based on Three-Dimensional Data
Fragment książki (Rozdział monografii pokonferencyjnej)
MNiSW
20
Poziom I
| Status: | |
| Autorzy: | Skublewska-Paszkowska Maria, Łukasik Edyta, Szydłowski Bartłomiej, Smołka Jakub, Powroźnik Paweł |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Arkusze wydawnicze: | 0,7 |
| Język: | angielski |
| Strony: | 146 - 155 |
| Scopus® Cytowania: | 2 |
| Bazy: | Scopus |
| Efekt badań statutowych | TAK |
| Materiał konferencyjny: | NIE |
| Publikacja OA: | NIE |
| Abstrakty: | angielski |
| The article explores the possibilities of using convolutional neural networks to recognize the type of tennis shots. The study compares two architectures of such networks: Inception-v3 and MobileNet. Two datasets consisting of images of tennis players making an impact are analyzed in order to identify the type and phase of impact. The images in each of the collections were assigned to the respective five following classes: backhand preparation phase, backhand shot, forehand preparation phase, forehand shot and non-shot. In each of them there is a tennis player silhouette and a racket skeleton consisting of markers. Images of tennis players are generated from motion recordings made using motion capture technology. The networks were trained with different values of the learning rate. In addition, the network training time and match results for the best-trained models are presented. In view of the parameters considered, the MobileNet network has proved to be a better model |
