Video Motion Analysis of Repetitive Exertions

描述（由申请人提供）：上肢肌肉骨骼损伤是手部高强度工作中常见的，包括高度重复的运动和努力。以前的接触评估方法包括使用工人手上或手臂上的仪器进行直接测量，或间接观察。然而，这两种方法在职业卫生实践中都是非常不切实际的。与仪器相比，间接观测缺乏精密度和准确性，不适合长观测周期，需要大量的分析时间。另外，在工作的手上安装传感器耗时太长，而且传感器可能会干扰正常的工作操作。我们正在开发一种新颖的实时视频处理算法，以便从传统的单摄像机数字视频中自动、不显眼地测量上肢运动。在这项探索性研究中，我们提出了一种视频特征提取方法，通过无标记视频运动分析来间接量化手部活动。我们正在与NIOSH全行业研究部门合作，该部门正在制作700多个视频，涉及三个研究地点的484名工人，他们的相关观察分析和纵向健康结果数据可供我们用于初步研究。实时

DESCRIPTION (provided by applicant): Upper extremity musculoskeletal injuries are common in hand intensive work involving highly repetitive motions and exertions. Previous exposure assessment methods involve either direct measurements using instruments attached to a worker's hands or arms, or indirect observations. Both methods however are highly impractical for occupational health practice. Compared to instruments, indirect observation lacks precision and accuracy, is not suitable for long observation periods, and requires considerable analyst time. Alternatively, attaching sensors on working hands is too time consuming, and sensors may interfere with normal working operation. We are developing novel real- time video processing algorithms to automatically and unobtrusively measure upper limb kinematics from conventional single camera digital video. In this exploratory research, we propose using a video feature extraction method to indirectly quantify hand activity using markerless video motion analysis. We are collaborating with the NIOSH Industry-wide Studies Branch, which is making more than 700 videos of 484 workers at three study locations, their associated observational analyses and longitudinal health outcome data available to us for this initial study. The real time video algorithms will first be developed and designed to be robust using sequential Bayesian estimation to track the motion of a general region of interest (ROI) on the upper extremities selected by the camera operator or analyst, and statistically estimating its velocity and acceleration. Video derived ROI kinematics from varying vantage points will then be compared against ground truth measurements obtained using a 3D infrared motion capture system, and conventional observational measurements of laboratory participants performing controlled repetitive motion tasks. We will also perform secondary analysis of videos from actual workplace activities taken and already studied by NIOSH, for algorithm refinement and for studying the correlation between the kinematic data and conventional observational exposure measures. We will evaluate the risk of an injury from the NIOSH health outcome data using our new exposure analysis method and compare the resulting dose-response model using the measured frequency, duty cycle, velocity and acceleration from our automatic video analysis method against conventional manual analysis in order to evaluate improvement over existing methods. This exploratory research could ultimately lead to a full-scale epidemiology and validation study of our video exposure assessment methodology that would leverage the existing NIOSH consortium database from seven participating laboratories in order to establish a model relating video derived kinematics and health outcome. The long-term goal is to develop a video-based direct reading exposure assessment instrument for upper limb repetitive motion activities that should be useful for evaluating the risk of injuries in the workplace, and for primry prevention.