VISION-BASED LIFTING LOAD ESTIMATION FOR PREVENTING LIFTING INJURIES

<div>Heavy and repetitive lifting tasks are commonly observed across many industries; however, the poor ergonomics of these tasks contribute to work-related musculoskeletal injuries</div><div>worldwide. Identifying when these tasks increase injury risks is essential for reducing workplace injuries. Current injury risk assessment tools require trained ergonomists to measure</div><div>worker posture, task repetition, and force exertion. While repetition and posture are easily</div><div>observable, the level of force exerted by the worker remains difficult to estimate without intrusive measurement techniques such as surface Electromyography(sEMG) sensors. In study</div><div>A, a video-based method for lifting risk estimation that can measure injury risks due to varying force levels without the need for intrusive sensors is proposed. The proposed method is</div><div>demonstrated with lifting tasks commonly observed in the workplace. The method consists</div><div>of a novel set of computer vision algorithms that monitor workers’ body motion, posture, and</div><div>facial expressions using only videos capturing the lifts. Extracted features were normalized</div><div>and used by machine learning models for classifying safety and risk levels determined by validated metrics of injury risk, i.e., lifting index and perceived physical effort (Borg scale). In</div><div>addition, this study discovered novel lifting risk indicators by investigating the relationships</div><div>between extracted features and lifting risks through interpretable machine learning and statistical inference techniques. In study B, a prototype decision support system that aims to</div><div>help people perform lifting risk assessment is developed. The proposed system implements</div><div>the video-based method from study A. A usability study is conducted to investigate the effect</div><div>of the decision support system on user performance and confidence, and demonstrates the</div><div>effectiveness of the proposed system. In summary, this thesis (a) proposes a non-intrusive</div><div>method for lifting risk assessment, (b) discovers novel lifting risk indicators, and (c) develops</div><div>a decision support system for helping people perform lifting risk assessment.</div>