Skeleton-Based Muscle Activity Prediction During Manual Material Handling

Electromyography (EMG) signals, which measure muscle electrical activity, are used to assess muscle function, activity, and fatigue levels during motion. However, collecting EMG data typically requires invasive or non-invasive sensors in controlled environments. This paper proposes a model to predict muscle activity levels from videos of people performing actions in real-world settings, enabling workplace applications. EMG signals and corresponding videos were collected from seven sites on the right side of subjects' bodies performing three actions with two weights. After preprocessing, muscle activity levels and joint coordinates were extracted to predict muscle activity levels.