Scholarlink

<p>The standardization and stability of sports postures are important factors to improve performance and reduce energy consumption in athletics. In this paper, we propose a posture correction method that integrates IMU (Inertial Measurement Unit) multimodal features and deep learning to address the problems of unstable detection of key points, accumulation of misjudgments, and lack of energy optimization in high-dynamic running scenarios. The method adopts thigh, calf and foot IMU modules to collect angle, acceleration and angular velocity signals, and combines YOLOv5 + P-RNet visual network to extract the posture key points and realize high-precision posture recognition. To solve the misjudgment problem, a correction algorithm based on motion coherence is designed to detect gait cycle anomalies and correct them through 1 + 2 and 2 + 1 window models. Meanwhile, energy economy analysis was introduced to compare low energy consumption (high efficiency) with high energy consumption (low efficiency) posture characteristics, and key kinetic parameters (ankle plantarflexion angular velocity, hip power output) were optimized during the correction process. The experiments validate the effectiveness of the method based on 1794 gait cycles: the accuracy of the correction is improved from 96.5 to 99.6%, and the performance is stable in fatigue state and high dynamic scenarios.</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

Exploration of improving motion posture correction in track and field events using deep learning

  • 期刊论文

  • 发表日期：

    2026年4月12日

查看全文
  • Lingyu Zhang,
  • Yang Yu,
  • Chenguang Han

摘要

The standardization and stability of sports postures are important factors to improve performance and reduce energy consumption in athletics. In this paper, we propose a posture correction method that integrates IMU (Inertial Measurement Unit) multimodal features and deep learning to address the problems of unstable detection of key points, accumulation of misjudgments, and lack of energy optimization in high-dynamic running scenarios. The method adopts thigh, calf and foot IMU modules to collect angle, acceleration and angular velocity signals, and combines YOLOv5 + P-RNet visual network to extract the posture key points and realize high-precision posture recognition. To solve the misjudgment problem, a correction algorithm based on motion coherence is designed to detect gait cycle anomalies and correct them through 1 + 2 and 2 + 1 window models. Meanwhile, energy economy analysis was introduced to compare low energy consumption (high efficiency) with high energy consumption (low efficiency) posture characteristics, and key kinetic parameters (ankle plantarflexion angular velocity, hip power output) were optimized during the correction process. The experiments validate the effectiveness of the method based on 1794 gait cycles: the accuracy of the correction is improved from 96.5 to 99.6%, and the performance is stable in fatigue state and high dynamic scenarios.