Predicting Weight and Strenuousness from High-Speed Videos of Subjects Attempting Lift

Intelligent understanding of human motions during lifting minimizes overexertion injuries by offering continuous monitoring and early intervention for people attempting heavy lifts at home or work. Standardized lift assessment methods such as the revised National Institute for Occupational Safety and Health lift equation require physical therapy expertise and lack subject perceptions of strenuousness in attempting lift tasks. We provide one of the first approaches to perform non-intrusive automated detection of strenuousness, weight lifted, and subject’s knowledge of weight using joint motions of subjects from multi-view high-speed color videos as subjects lift varying weights, with and without prior weight knowledge. We show average accuracies of 81.32% and 77.23% by using convolutional neural networks to automatically detect low versus high weight and strenuousness. Our work informs monitoring technologies for individuals engaged in heavy lifting in manufacturing, warehousing, nursing, and emergency response.