Recognition and Prediction of Human Actions in Industrial Human-Robot Collaboration Scenarios

This paper focuses on human action recognition in human-robot collaboration (HRC) scenarios, specifically investigating human motion recognition and human motion prediction in industrial environments. It employs a 3D convolutional neural network (3D CNN) for action recognition and integrates Transformer, LSTM, and DyHead mechanisms to construct a human motion prediction model, enabling the prediction of future human movements. The system collects video data from a simulated wire harness assembly scene to enhance the robotic arm's dynamic obstacle avoidance capabilities and ensure operational safety in industrial settings. Experimental results demonstrate that the system effectively improves safety and operational efficiency in complex HRC scenarios.