Phase Segmentation and Phase-Specific Kinematic Feature Extraction of Hurdle Clearance Based on Monocular Video and Markerless Pose Estimation.

Hurdle technique analysis requires accurate identification of key phases and kinematic features, but conventional biomechanical methods are often costly, equipment-dependent, and difficult to apply in front-line training. This study developed a low-cost monocular-video-based framework for rapid hurdle clearance analysis in practical training settings. Thirty-seven physical education college students with different hurdling skill levels were recruited as participants, and side-view videos of their hurdle clearance were recorded. The proposed pipeline combined YOLO26 hurdle detection, RTMPose markerless pose estimation, rule-based key-event detection, phase segmentation, and phase-specific kinematic feature extraction. The results showed that the hurdle detection model achieved high accuracy, with bounding-box mAP@0.5 of 0.992 and mask mAP@0.5 of 0.971. Pose estimation showed good agreement with manual annotations, with an overall RMSE of 8.25 px and PCK of 97.64%. The rule-based phase segmentation method achieved an overall event localization MAE of 0.74 frames and RMSE of 1.55 frames, outperforming LSTM and TCN temporal baselines. Core distance and most angle variables also showed high agreement with manually recalculated values. These findings indicate that monocular video and markerless pose estimation can provide an accurate, low-cost, and practical tool for hurdle phase segmentation and kinematic assessment in routine training contexts.