Scholarlink

<p>Progressive walking fatigue alters gait kinematics, yet low-cost, physiologically anchored continuous monitoring remains underdeveloped. This study aimed to identify a parsimonious set of Kinect-derived stride features capable of faithfully reconstructing a physiologically validated fatigue trajectory on a per-participant basis. Thirty-three healthy young adults walked self-paced to volitional fatigue on a treadmill while a Kinect v2 sensor recorded stride-level joint kinematics. Borg RPE served as the universal per-participant optimization target, DTW-normalized to a common 0–100% Fatigue Progression Time axis; tibialis-anterior sEMG-MPF (<i>n</i> = 10 subset) provided independent group-level validation confirming the near-linear trajectory of walking-induced fatigue. Grey Wolf Optimization (GWO; 10 runs × 50 iterations) ranked 16 clinician-recommended features in a 16-dimensional discovery stage; ablation analysis identified a clear performance plateau at <i>N</i> = 8. The eight-feature model (hip sagittal angle; right/left knee angles and angular velocities; step height; spine angle; toe-out) achieved mean MAE = 6.76 ± 5.61% points (95% CI 5.39–8.92), RMSE = 8.87 ± 6.24 (95% CI 7.26–11.46), and Pearson <i>r</i> = 0.965 (95% CI 0.954–0.973), corresponding to a mean percentage-based reconstruction accuracy of 93.2%. Per-participant linear coefficients were fully interpretable, with hip sagittal angle emerging as the dominant contributor in 57% of participants. This in-sample reconstruction study establishes a transparent, data-efficient feature-selection foundation for future out-of-sample fatigue prediction, pending validation in overground and clinical settings.</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

Evaluation of Fatigue Intensity from Gait Data: A Mathematical Model

  • 期刊论文

  • 发表日期：

    2026年7月7日

查看全文
  • Parham Yazdani,
  • Hadi Soltanizadeh,
  • Ali Maleki,
  • Farhad Azadi,
  • Mohammad Zahraei,
  • Shib Sankar Sana

摘要

Progressive walking fatigue alters gait kinematics, yet low-cost, physiologically anchored continuous monitoring remains underdeveloped. This study aimed to identify a parsimonious set of Kinect-derived stride features capable of faithfully reconstructing a physiologically validated fatigue trajectory on a per-participant basis. Thirty-three healthy young adults walked self-paced to volitional fatigue on a treadmill while a Kinect v2 sensor recorded stride-level joint kinematics. Borg RPE served as the universal per-participant optimization target, DTW-normalized to a common 0–100% Fatigue Progression Time axis; tibialis-anterior sEMG-MPF (n = 10 subset) provided independent group-level validation confirming the near-linear trajectory of walking-induced fatigue. Grey Wolf Optimization (GWO; 10 runs × 50 iterations) ranked 16 clinician-recommended features in a 16-dimensional discovery stage; ablation analysis identified a clear performance plateau at N = 8. The eight-feature model (hip sagittal angle; right/left knee angles and angular velocities; step height; spine angle; toe-out) achieved mean MAE = 6.76 ± 5.61% points (95% CI 5.39–8.92), RMSE = 8.87 ± 6.24 (95% CI 7.26–11.46), and Pearson r = 0.965 (95% CI 0.954–0.973), corresponding to a mean percentage-based reconstruction accuracy of 93.2%. Per-participant linear coefficients were fully interpretable, with hip sagittal angle emerging as the dominant contributor in 57% of participants. This in-sample reconstruction study establishes a transparent, data-efficient feature-selection foundation for future out-of-sample fatigue prediction, pending validation in overground and clinical settings.