Real-Time Framework for Talent Swimmer Detection | ||
| International Journal of Applied Intelligent Computing and Informatics | ||
| Article 3, Volume 1, Issue 2, September 2025, Pages 36-45 PDF (1.02 M) | ||
| Document Type: Original Article | ||
| DOI: 10.21608/ijaici.2025.414064.1015 | ||
| Authors | ||
| Hossam Fakher* 1; Alsayed badr2; Ahmed Hassanein3; Sara Sweidan4 | ||
| 1Department of Artificial Intelligence, Faculty of Computers and Artificial Intelligence, Benha University, Benha, Egypt | ||
| 2Department of Information Systems, College of Information Technology, Misr University for Science & Technology, Giza, Egypt. Department of Scientific Computing, Faculty of Computer and Artificial Intelligence, Benha University, Benha, Egypt; | ||
| 3Department of Kinesiology, Specifications Biomechanics, Faculty of Physical Education, Dumyat University; | ||
| 4Department of Artificial Intelligence, Faculty of Computers and Artificial Intelligence, Benha University, Benha 13518, Egypt. Faculty of Computer Science and Engineering, New Mansoura University, New Mansoura, Egypt; | ||
| Abstract | ||
| This study presents a real-time framework for swimmer talent identification that integrates state-of-the-art pose estimation and machine learning classification techniques. To address the limitations of traditional pose estimation methods in aquatic environments, RTMPose is employed to extract reliable 2D joint keypoints. Temporal consistency across sequences is achieved using the RIFE interpolation model, selected for its efficiency in standardizing frame counts while avoiding the computational overhead of temporal deep learning models such as LSTMs or 3D CNNs. The dataset, consisting of underwater breaststroke footage, was augmented and balanced using SMOTE, with sensitivity analysis highlighting both its benefits for minority classes and the risk of overfitting. A comprehensive evaluation of twelve classifiers demonstrated that ensemble methods, particularly LightGBM, achieved superior results, yielding a cross-validation F1 score of 93.6% and a test F1 score of 96.8%. While the framework shows strong promise for practical use in sports analytics, its current evaluation is limited to breaststroke and underwater footage. Future work will expand to multiple swimming styles, above-water perspectives, and diverse pool environments to ensure broader generalization. | ||
| Keywords | ||
| Object Detection; Pose estimation; RTMPose; Swimmer; Talent and Machine Learning | ||
|
Statistics Article View: 37 PDF Download: 44 |
||
