The present invention proposes a skeleton-based industrial site monitoring method that predicts a worker's movement path through skeleton analysis based on video footage captured by cameras installed throughout the industrial site, thereby preventing path overlap with a robot. The method includes the following steps: generating skeleton data by recognizing joint points of a worker based on a captured video of the industrial site where a robot is present, estimating an expected movement path of the worker based on the generated skeleton data, searching for an optimal path for the robot to avoid collision with the worker based on the estimated expected movement path of the worker, and controlling the robot according to the searched optimal path.