Scholarlink

Abstract <p>This research introduces a multimodal framework for in-depth gait analysis and evaluation of hip joint asymmetry across different walking speeds in healthy young Indian adults (18-25 years old). The gait data were obtained from 55 subjects using synchronized RGB-D cameras and wearable IMUs located in frontal and sagittal planes at four treadmill speeds (1, 3, 5, and 7 mph), thereby offering extensive spatiotemporal and kinematic data. A Hybrid Deep Networks Ensemble (EHDN) that combines CNN, LSTM-CNN, and CNN-LSTM architectures via a bagging strategy was made to infer walking speed from RGB-D data. The raised ensemble secured the classifying accuracy of 95.96%, hence surpassing the individual deep learning models. At the same time, IMU-based hip asymmetry analysis through two-way ANOVA showed significant main and interaction effects of walking speed and hip side, with the right hip showing the largest variance across speeds, thus indicating a mild, speed-dependent asymmetry even among the healthy individuals. The proposed multimodal, low-cost, and non-intrusive framework holds great promise for clinical gait diagnostics, rehabilitation, and real-time edge deployment. The results not only contribute to the development of intelligent and interpretable gait assessment solutions that are accessible for both research and applied healthcare environments but also steer such advancements in the right direction.</p> Graphical abstract <p></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

EDL-HJAE: An ensemble deep learning framework for multimodal gait analysis and hip joint asymmetry evaluation

  • 期刊论文

  • 发表日期：

    2026年3月17日

查看全文
  • Vishwanath Bijalwan,
  • Satyam Mishra,
  • Abdul Mannan Khan,
  • Hangyeol Baek,
  • Venkatesan Vellaiyan,
  • Youngshik Kim

摘要

Abstract

This research introduces a multimodal framework for in-depth gait analysis and evaluation of hip joint asymmetry across different walking speeds in healthy young Indian adults (18-25 years old). The gait data were obtained from 55 subjects using synchronized RGB-D cameras and wearable IMUs located in frontal and sagittal planes at four treadmill speeds (1, 3, 5, and 7 mph), thereby offering extensive spatiotemporal and kinematic data. A Hybrid Deep Networks Ensemble (EHDN) that combines CNN, LSTM-CNN, and CNN-LSTM architectures via a bagging strategy was made to infer walking speed from RGB-D data. The raised ensemble secured the classifying accuracy of 95.96%, hence surpassing the individual deep learning models. At the same time, IMU-based hip asymmetry analysis through two-way ANOVA showed significant main and interaction effects of walking speed and hip side, with the right hip showing the largest variance across speeds, thus indicating a mild, speed-dependent asymmetry even among the healthy individuals. The proposed multimodal, low-cost, and non-intrusive framework holds great promise for clinical gait diagnostics, rehabilitation, and real-time edge deployment. The results not only contribute to the development of intelligent and interpretable gait assessment solutions that are accessible for both research and applied healthcare environments but also steer such advancements in the right direction.

Graphical abstract