Phase-aware kinematic measurement of standing long jump based on human posture vision

The standing long jump (SLJ) is widely used in large-scale physical fitness assessment. However, routine SLJ testing focuses primarily on jump distance rather than providing technique-oriented feedback for skill improvement. In this paper, we propose a phase-aware Kinematic SLJ measurement method based on Human Posture Vision (KSHPV). By making use of monocular side-view video to construct SLJ dataset, KSHPV formulates a lightweight detection-and-pose framework to extract 2D skeletal trajectories. A temporal key-event localization algorithm is designed to identify take-off, peak flight, and landing for phase alignment across trials. Based on the aligned phases, KSHPV further constructs a set of interpretable kinematic indicators that describe countermovement preparation, propulsion, upper–lower limb coordination, flight timing, and landing trunk control. KSHPV introduces a confidence-aware quality-control strategy to improve feature stability and reduce unreliable measurements in routine field recordings. The extraction of standardized phase-aware descriptors supports downstream performance modeling and hypothesis-driven analysis. By using the resulting phase-aligned indicators under participant-grouped cross-validation (204 participants; 219 analyzed trials), KSHPV achieves a mean absolute error (MAE) of 17.25 ± 2.67 cm and R² of 0.665 ± 0.068 with a random forest regressor, substantially outperforming an anthropometrics-only baseline (MAE 28.92 ± 2.43 cm; R² 0.160 ± 0.091). Hypothesis-driven analyses confirm that phase alignment markedly improves explanatory power (ΔR² = 0.289), while coordination features provide a modest but consistent additional gain (ΔR² = 0.012). Robustness analyses across gender/BMI strata and simulated noise/missingness quantify the operating boundary of monocular pose-based assessment. KSHPV and its indicator set enable scalable, interpretable technique analysis for SLJ in real-world testing and teaching contexts.