(英) |
Detection of fast-moving shuttlecocks is essential for badminton video analysis. Several methods based on deep learning have been proposed in the literature. TrackNetV2 is a state-of-the-art shuttlecock detection model which uses the U-Net architecture, but we believe there is room for further performance improvements. In this research, we extend TrackNetV2 via employing deep residual learning. Experimental results on a public shuttlecock detection dataset demonstrates that our proposed method performs better than the original TrackNetV2 with respect to precision, recall, F1 score and accuracy. |