Scholarlink

<p>Olympic-style weightlifting requires a lot of coordination and complex motions, and keeping track of the barbell accurately is important for both performance evaluation and lowering the risk of injury. This work presents a single-camera, computer-vision framework that corrects for perspective distortion, automatically localizes the barbell using a YOLO-based detector, then tracks the barbell center with multiple 2D trackers, and finally applies a rule-based classifier to categorize snatch trajectories into four established types. The original rule-based classifier achieved 70% accuracy on a dataset of 10 competition snatch videos (about 6000 frames). The proposed YOLO initialization only changed the mean trajectory deviations by a few centimeters compared to manually initialized tracks, and the score category stayed the same for most lifts. Eight barbell kinematic variables are extracted, and three spatial measures–vertical peak height <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(Y_{\text {max}}\)</EquationSource> </InlineEquation>, starting horizontal setup <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(X_1\)</EquationSource> </InlineEquation>, and bar drop <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(Y_{\text {drop}}\)</EquationSource> </InlineEquation>–are integrated into a 0–4 performance score matched to five qualitative categories from “Very Bad” to “Excellent.” Sensitivity evaluations demonstrate that modest tracking noise and ±5–10% calibration perturbations infrequently affect the conformity of these kinematic variables to standard reference ranges, suggesting that the score is resilient to practical video and calibration inaccuracies. Further assessments of velocity, power, and tracker jitter quantify the correlation between dynamic barbell features and algorithmic decisions, as well as their influence on lift quality, yielding a more biomechanically interpretable and numerically reliable instrument for assessing applied snatch technique.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Single-Camera Barbell Trajectory Analysis for the Snatch Lift

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Dhairya Shah,
  • Hetav Raval,
  • Christopher Taber,
  • Tolga Kaya,
  • Eva Maddox,
  • Mehul S. Raval

摘要

Olympic-style weightlifting requires a lot of coordination and complex motions, and keeping track of the barbell accurately is important for both performance evaluation and lowering the risk of injury. This work presents a single-camera, computer-vision framework that corrects for perspective distortion, automatically localizes the barbell using a YOLO-based detector, then tracks the barbell center with multiple 2D trackers, and finally applies a rule-based classifier to categorize snatch trajectories into four established types. The original rule-based classifier achieved 70% accuracy on a dataset of 10 competition snatch videos (about 6000 frames). The proposed YOLO initialization only changed the mean trajectory deviations by a few centimeters compared to manually initialized tracks, and the score category stayed the same for most lifts. Eight barbell kinematic variables are extracted, and three spatial measures–vertical peak height \(Y_{\text {max}}\) , starting horizontal setup \(X_1\) , and bar drop \(Y_{\text {drop}}\) –are integrated into a 0–4 performance score matched to five qualitative categories from “Very Bad” to “Excellent.” Sensitivity evaluations demonstrate that modest tracking noise and ±5–10% calibration perturbations infrequently affect the conformity of these kinematic variables to standard reference ranges, suggesting that the score is resilient to practical video and calibration inaccuracies. Further assessments of velocity, power, and tracker jitter quantify the correlation between dynamic barbell features and algorithmic decisions, as well as their influence on lift quality, yielding a more biomechanically interpretable and numerically reliable instrument for assessing applied snatch technique.