A Computer Vision Lifting Monitor

项目摘要/摘要 重复的手动起重是一个重要的职业健康和安全问题，在 仓储、配送中心、包裹递送、运输和精益制造。这些类型的 从人体工程学的角度分析任务是最具挑战性的，特别是在多任务情况下 举起不同的物品发生在许多地点，涉及整个工作日不同的身体姿势。 对于工业来说，手动测量分析所需的参数是具有挑战性和资源密集型的 今天的从业者。这项研究的首要目标是为起重建立一个计算机视觉风险模型， 将其集成到样机中，并与常规RNLE进行了现场比较 方法：研究方法。自动化工作分析可能会提供更客观、准确、可重复和高效的曝光 评估工具，而不是传统的观测方法。此外，它还提供了方便的量化 其他曝光变量，包括起重运动学(即速度和加速度)个体差异， 和姿势；适用于长期、直接阅读暴露评估；并提供动画数据 可视化与视频同步，以确定干预措施。这项研究翻译了已经收集的 来自具有里程碑意义的计算机前瞻性研究数据库的工作和相应健康结果的视频 视力下腰痛风险评估。它利用了庞大的视频数据库和相应的曝光 772名工人执行的升降活动(即子任务)的测量和健康数据 三项队列研究，由我们在NIOSH、犹他大学和华盛顿大学的研究伙伴收集 威斯康星--密尔沃基。他们是由NIOSH资助的多机构美国实验室财团的一部分 最近，在一项关于下腰痛的前瞻性流行病学研究中，对各种行业的工人进行了研究。 该联盟的视频将通过提取新的视频特写曝光措施进行分析，包括提升 姿势、躯干和载荷运动学。视频曝光评估数据将与财团合并 观察性暴露措施和健康结果数据。我们将检验添加计算机的假设 具有联合暴露变量的视觉暴露变量可以提高预测下背部的性能 疼痛。该项目将完善和编程视频暴露评估算法的姿势分类，躯干 角度和躯干，并将运动学加载到原型装置中。新的曝光算法将在 选择工业场地，并与传统观测方法进行一致性和实用性比较 (R2P)。此翻译研究提供了一个前所未有的机会来利用独特的视频和关联 已经收集的暴露和健康结果数据，结合量化的新技术 曝光。这项研究涉及制造业，以及运输、仓储和公用事业 各部门之间的合作以及肌肉骨骼健康跨部门议程。

Project Summary/ Abstract Repetitive manual lifting is a significant occupational health and safety concern and is highly prevalent in warehousing, distribution centers, package delivery, transportation, and lean manufacturing. These types of tasks are the most challenging to analyze from an ergonomics perspective, particularly in multi-task situations where lifting varied items occurs in numerous locations, involving variable body postures throughout the workday. Manually measuring the parameters needed for analysis is challenging and resource intensive for industry practitioners today. The overarching goal of this research is to create a computer vision risk model for lifting, incorporate it into a prototype instrument, and field evaluate the instrument in comparison to conventional RNLE methods. Automated job analysis potentially offers a more objective, accurate, repeatable, and efficient exposure assessment tool than conventional observational methods. Furthermore, it provides convenient quantification of additional exposure variables, including lifting kinematics (i.e., speed and acceleration) individual differences, and postures; is suitable for long-term, direct reading exposure assessment; and offers animated data visualization synchronized with video for identifying interventions. This research translates already collected videos of jobs and corresponding health outcomes from a landmark prospective study database for computer vision lower back pain risk assessment. It leverages the vast database of videos and corresponding exposure measures and health data for lifting and lowering activities (i.e., subtasks) performed by 772 workers across the three cohort studies, collected by our study partners at NIOSH, the University of Utah, and the University of Wisconsin-Milwaukee. They are part of a multi-institutional NIOSH funded consortium of U.S. laboratories that recently studied workers in a wide variety of industries in a prospective epidemiology study on lower back pain. The consortium videos will be analyzed by extracting the new video feature exposure measures, including lifting postures, and torso and load kinematics. The video exposure assessment data will be combined with consortium observational exposure measures and health outcome data. We will test the hypothesis that adding computer vision exposure variables with consortium exposure variables can enhance performance of predicting lower back pain. This project will refine and program video exposure assessment algorithms for posture classification, torso angle and trunk and load kinematics into a prototype device. The new exposure algorithms will be tested in selected industrial sites and compared against conventional observational methods for consistency and utility (r2p). This translational research offers an unprecedented opportunity to exploit unique videos and associated exposure and health outcome data already collected, in combination with new technology for quantifying exposures. This research addresses the manufacturing, and the transportation, warehousing, and utilities NORA sectors, as well as the musculoskeletal health cross sector agendas.