PFI: AIR - TT: Vision-based ergonomic risk assessment of working postures

这个PFI: AIR技术翻译项目的重点是翻译基于计算机视觉的姿势分类和姿势估计的基础研究成果，以满足快速简便的姿势人体工程学风险评估的需求。这适用于从事体力劳动的职业工人，他们的工作姿势笨拙，因此患有与工作有关的肌肉骨骼疾病（WMSDs）。翻译后的技术很重要，因为它使从业者能够在实际伤害之前识别潜在的WMSD风险，从而最大限度地减少潜在的伤害风险。鉴于美国WMSD病例数量庞大（例如，每年超过36万例）和相关的巨额工人赔偿费用（例如，每年超过142亿美元），本研究有助于发展和维持安全、健康和富有成效的劳动力队伍，这将对社会产生改善性影响。该项目将产生一个功能齐全的基于视觉的人体工程学风险评估软件原型，它具有以下独特的特点：快速（例如，接近实时地处理视频帧），简单（例如，不需要培训来评估人体工程学风险）和自动化（例如，不需要手动风险评估）。与目前人体工程学行业中基于人工观察的风险评估相比，这些功能使从业者可以节省大量的时间和精力进行人体工程学风险评估。该项目解决了以下技术差距，因为它从研究发现转化为商业应用。首先，基于跟踪的姿态估计是估计姿态严重程度的基本组成部分，它提供了视频帧的快速处理，但需要在没有目标丢失的情况下保持鲁棒性，而目标丢失可能由于真实工作场所频繁的遮挡和环境噪声而发生。在本研究中，基于跟踪的方法将与基于检测的姿态估计相结合，并采用不同的特征，如刚处理的空间信息和参考帧。一旦原型开发出来，它将在不同的实际工作场所进行测试，例如制造装配线，以确保其稳健性和可用性。此外，参与该项目的人员，包括研究生和本科生，将通过与人体工程学和商业专家以及潜在客户的互动（例如，客户互动，市场研究和现场验证）以及参与密歇根大学创业中心提供的活动（例如，创业课程和研讨会）来获得创业经验。该项目与北美最大的人体工程学咨询公司Humantech、美国汽车、航空航天和农业机械工人联合会（UAW）以及一家风险资本家合作，提供真实世界的测试环境，并指导这项技术将研究成果转化为商业现实的商业化方面。

This PFI: AIR Technology Translation project focuses on translating fundamental research findings in computer vision-based posture classification and pose estimation to fill the need for rapid and easy postural ergonomic risk assessment. This is applicable for occupational workers involved in physically demanding labor with awkward working postures, and who consequently suffer from work-related musculoskeletal disorders (WMSDs). The translated technology is important because it enables practitioners to identify potential WMSD risks in advance of actual injury, thus minimizing the potential injury risks. Given the massive number of WMSD cases (e.g., over 360,000 cases per year) and huge associated workers' compensation costs (e.g., over $14.2 billion per year) in the U.S., this research contributes to the development and maintenance of a safe, healthy, and productive workforce, which will have an ameliorative impact on society. The project will result in a fully-functioning software prototype of vision-based ergonomic risk assessment, which has the following unique features: rapid (e.g., near real-time to process video frames), easy (e.g., no training to assess ergonomic risks is needed), and automated (e.g., no need for manual risk assessment). These features allow practitioners to save a significant amount of time and effort on ergonomic risk assessment, when compared to the current manual observation-based risk assessment in the ergonomic industry. This project addresses the following technology gaps as it translates from research discovery toward commercial application. First, a tracking-based pose estimation, which is a fundamental component to estimate posture severity, provides fast processing of video frames but needs to be robust without target loss, which can happen due to frequent occlusions and environmental noises in real worksites. In this research, the tracking-based approach will be integrated with a detection-based pose estimation along with the adoptions of different features like spatial information and reference frames from just processed ones. Once the prototype is developed, it will be tested in diverse real worksites such as manufacturing assembly lines to ensure its robustness and usability. In addition, personnel involved in this project, including graduate and undergraduate students, will receive entrepreneurship experience through interaction with ergonomic and business experts and potential customers (e.g., customer interaction, market study, and field validation) as well as engagement in the activities offered from the University of Michigan Center for Entrepreneurship (e.g., entrepreneurship classes and workshops). The project engages Humantech, the largest ergonomic consulting firm in North America, UAW (United Automobile, Aerospace and Agricultural Implement workers of America), and a venture capitalist, to provide real-world test environment and guide commercialization aspects in this technology translation effort from research findings toward commercial reality.