The standing long jump (SLJ) is widely used for large-scale fitness assessment, yet existing distance measurement solutions remain labor-intensive or hardware-dependent. We frame SLJ distance estimation as a markerless monocular imaging-metrology problem and present a reusable vision pipeline comprising key-frame selection, person detection, semantic segmentation of the heel and jump-mat, homography-based plane mapping, and distance computation with polynomial compensation for perspective-edge bias. Field tests with a high-frame-rate camera demonstrate real-time performance (≈ 23 FPS, ~ 42.7 ms per frame) and centimeter-level accuracy. Using manual tape measurement as the reference, the overall system attains an MAE of 0.71 cm; ablations show that removing key modules sharply degrades accuracy (e.g., 25.07 cm without incomplete-athlete handling; 4.35 cm without single-view perspective mapping; 2.31 cm without convex-hull/curvature-based heel extraction). Minimal calibration files and inference scripts are provided to support reproducibility and deployment in school testing.
Kuang et al. (Mon,) studied this question.