Human Activity Recognition in MMH: Pilot Study on Lifting and Lowering Classification

Work-related musculoskeletal disorders are a leading cause of industrial injuries, with manual material handling (MMH) contributing to over 500,000 cases annually in the U.S. This study supports the development of a human digital twin (HDT) by enabling accurate human activity recognition for meaningful analysis. Using computer vision-based OpenPose for pose estimation, we classify two of fundamental MMH moves, lifting and lowering with Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM) models. LSTM proved more robust, achieving 92.2% accuracy. This approach demonstrates the feasibility of non-invasive MMH monitoring, enabling real-time safety assessments and enhancing HDT-based industrial applications.