Video-Based Rowing Stability Measurement by Self-Learning Strategy

Introduction: Rowing requires high coordination and consistency of movements. Stable and consistent rowing strokes are crucial for performance; however, methods to evaluate the rowing stability of players in the sport have not been thoroughly addressed. Methods: This study analyzes rowing movements recorded from a rowing machine to measure the stability of individual players. Joint points were extracted using MediaPipe and segmented into rowing cycles using the proposed Rowing Cycle Segmentation (RSC) method. Each cycle’s joint data was then input into the Pose Temporal Cycle-Consistency (PTCC) network, which is based on temporal cycle consistency learning, to train the model. The PTCC network aligned cycle timelines and produced frame-by-frame retrieval results. The average number of matched frames across cycles was used as the stability measure. Results: The trained PTCC network successfully aligned the timelines of cycled videos and generated frame-by-frame retrieval results. The average matched frames among cycled videos were used to represent rowing stability. Experimental results demonstrated that the proposed method could effectively quantify rowing stability among players. Discussion: This study demonstrates the potential of non-contact, video-based approaches with self-learning strategies for objective evaluation of rowing stability. Compared to sensor-based methods, the framework minimizes interference while maintaining reliable accuracy. Conclusion: The combined RSC and PTCC framework provides an efficient and robust tool for assessing rowing stability. It enables clear differentiation of players’ consistency and identification of the most stable performer. Future work will focus on improving model robustness and validating its applicability in real-world competitive rowing environments.