Scholarlink

<p>Real-time rehabilitation assessment is essential to mitigate the shortage of physical therapists and to enable effective home-based recovery. Yet, existing systems mainly provide retrospective analysis, while current real-time approaches often lack the precision to detect subtle compensatory patterns. This work proposes a computer vision framework that leverages RGB cameras to assess exercise quality and deliver corrective guidance. Firstly, MediaPipe Pose is employed to capture 33 body landmarks in real time, followed by baseline establishment and height normalization for consistent analysis across diverse patients and environments. Then, a dual regulation mechanism is applied: static points maintain postural stability through temporal filtering that distinguishes normal adjustments from sustained positional deviations, while dynamic points ensure motion completion, trajectory compliance, and compensatory movement detection. Based on this regulation, the system delivers immediate multimodal feedback through synchronized visual animations, speech synthesis, and textual instructions. Validation on the REHAB24-6 dataset demonstrates that our frame-level analysis, unlike segment-level ground truth, detects 0.7% errors within correct actions and 1.92% correct frames within incorrect actions, enabling precise temporal deviation localization. This work provides an accurate and flexible solution for real-time rehabilitation monitoring, supporting patient adherence and extending applications to post-surgical recovery, rural healthcare delivery, and telerehabilitation.</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

Real-time rehabilitation assessment and corrective guidance driven by dual regulation pose analysis

  • 期刊论文

  • 发表日期：

    2026年3月18日

查看全文
  • Yi Qiao,
  • Zilong Wang

摘要

Real-time rehabilitation assessment is essential to mitigate the shortage of physical therapists and to enable effective home-based recovery. Yet, existing systems mainly provide retrospective analysis, while current real-time approaches often lack the precision to detect subtle compensatory patterns. This work proposes a computer vision framework that leverages RGB cameras to assess exercise quality and deliver corrective guidance. Firstly, MediaPipe Pose is employed to capture 33 body landmarks in real time, followed by baseline establishment and height normalization for consistent analysis across diverse patients and environments. Then, a dual regulation mechanism is applied: static points maintain postural stability through temporal filtering that distinguishes normal adjustments from sustained positional deviations, while dynamic points ensure motion completion, trajectory compliance, and compensatory movement detection. Based on this regulation, the system delivers immediate multimodal feedback through synchronized visual animations, speech synthesis, and textual instructions. Validation on the REHAB24-6 dataset demonstrates that our frame-level analysis, unlike segment-level ground truth, detects 0.7% errors within correct actions and 1.92% correct frames within incorrect actions, enabling precise temporal deviation localization. This work provides an accurate and flexible solution for real-time rehabilitation monitoring, supporting patient adherence and extending applications to post-surgical recovery, rural healthcare delivery, and telerehabilitation.