In order to compete more effectively in high level water sport of canoeing, it is essential for coaches and athletes to have a solid understanding of paddle motion. This paper presents a based-AI solution for precisely capturing the paddle trajectory when canoeing. Leveraging state-of-The-Art object detection, instance segmentation algorithm, YOLO, helps to detect the accurate shape of the paddle in the different frames of the video. We also offer a comparative study between two popular tracking algorithms: BoT-SORT and ByteTrack. Additionally, this work investigates the impact of the most commonly used optimizers in machine learning including SGD, Adam and AdamW on the system's overall performance. Finally, we found that BoT-SORT performed better than ByteTrack in following and recognizing the paddle in a higher quantity of frames. Moreover, in terms of training procedure, the results showed that SGD outperformed the two adaptive optimizers Adam and AdamW overall, with an average precision of 0.63 as opposed to 0.59 for both Adam and AdamW.
AI-Driven Paddle Motion Detection
Zanela A.
2024-01-01
Abstract
In order to compete more effectively in high level water sport of canoeing, it is essential for coaches and athletes to have a solid understanding of paddle motion. This paper presents a based-AI solution for precisely capturing the paddle trajectory when canoeing. Leveraging state-of-The-Art object detection, instance segmentation algorithm, YOLO, helps to detect the accurate shape of the paddle in the different frames of the video. We also offer a comparative study between two popular tracking algorithms: BoT-SORT and ByteTrack. Additionally, this work investigates the impact of the most commonly used optimizers in machine learning including SGD, Adam and AdamW on the system's overall performance. Finally, we found that BoT-SORT performed better than ByteTrack in following and recognizing the paddle in a higher quantity of frames. Moreover, in terms of training procedure, the results showed that SGD outperformed the two adaptive optimizers Adam and AdamW overall, with an average precision of 0.63 as opposed to 0.59 for both Adam and AdamW.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
