Abstract
<title>Abstract</title> <p>Video-based movement quality assessment offers low-cost feedback for physical education, but existing methods mainly predict overall scores without explaining phase-specific biomechanical deviations. This paper presents a phased assessment framework for the standing long jump, combining biomechanically guided standard motion modelling with a lightweight ST-GCN. From 200 participants (100 men, 100 women), 562 valid video trials were collected. MMPose extracted 2D keypoints, which were reconstructed into 3D skeleton sequences (Human3.6M, 17 joints) using VideoPose3D. A five-phase standard template (preparation, power generation, take-off, flight, landing stability) was built based on biomechanical evidence and expert knowledge, mapped into three scoring domains (take-off, flight, landing). A four-layer lightweight ST-GCN then regressed phase-specific and weighted total scores under expert supervision. The framework achieved high consistency with expert scores (Spearman ρ=0.94, RMSE=0.12, MAE=0.10) with only 0.48M parameters and 2.7 GFLOPs. It also provides interpretable feedback (e.g., insufficient push-off, low take-off angle, poor aerial posture). The proposed method offers efficient, low-cost, interpretable support for assessing standing long jump quality in school physical education and grassroots training.</p>