College of Electronic and Information Engineering, Tongji University, Shanghai 201804, China.
Frontiers Science Center for Intelligent Autonomous Systems, Shanghai 201210, China.
Sensors (Basel). 2023 Apr 8;23(8):3825. doi: 10.3390/s23083825.
With the changes in human work and lifestyle, the incidence of cervical spondylosis is increasing substantially, especially for adolescents. Cervical spine exercises are an important means to prevent and rehabilitate cervical spine diseases, but no mature unmanned evaluating and monitoring system for cervical spine rehabilitation training has been proposed. Patients often lack the guidance of a physician and are at risk of injury during the exercise process. In this paper, we first propose a cervical spine exercise assessment method based on a multi-task computer vision algorithm, which can replace physicians to guide patients to perform rehabilitation exercises and evaluations. The model based on the Mediapipe framework is set up to construct a face mesh and extract features to calculate the head pose angles in 3-DOF (three degrees of freedom). Then, the sequential angular velocity in 3-DOF is calculated based on the angle data acquired by the computer vision algorithm mentioned above. After that, the cervical vertebra rehabilitation evaluation system and index parameters are analyzed by data acquisition and experimental analysis of cervical vertebra exercises. A privacy encryption algorithm combining YOLOv5 and mosaic noise mixing with head posture information is proposed to protect the privacy of the patient's face. The results show that our algorithm has good repeatability and can effectively reflect the health status of the patient's cervical spine.
随着人类工作和生活方式的改变，颈椎病的发病率大幅上升，尤其在青少年中更为常见。颈椎运动是预防和康复颈椎疾病的重要手段，但目前还没有成熟的无人评估和监测系统来对颈椎康复训练进行评估。患者在进行锻炼时往往缺乏医生的指导，存在受伤的风险。在本文中，我们首先提出了一种基于多任务计算机视觉算法的颈椎运动评估方法，该方法可以替代医生来指导患者进行康复运动和评估。该模型基于 Mediapipe 框架构建，用于构建人脸网格并提取特征，以计算 3-DOF（三自由度）中的头部姿势角度。然后，根据上述计算机视觉算法获取的角度数据，计算 3-DOF 中的顺序角速度。之后，通过颈椎运动数据采集和实验分析，对颈椎康复评价系统和指标参数进行了分析。提出了一种结合 YOLOv5 和马赛克噪声混合与头部姿势信息的隐私加密算法，以保护患者面部的隐私。结果表明，我们的算法具有良好的可重复性，可以有效反映患者颈椎的健康状况。
