Action Transition Recognition Using Principal Component Analysis for Agricultural Robot Following

During harvesting using a robot, a farmer performs the actions of harvesting the crop and placing it on the robot. Meanwhile, the robot follows the farmer, and the farmer is comfortable if the distance at which the robot follows changes dependent on the farmer’s actions. In this study, we proposed a method of recognizing the action transition using the result of the principal component analysis of the skeletal information. It was confirmed that the proposed method recognized the transition to the placing action at the start of the action by the values of the 1st–3rd principal components. In addition, the proposed method recognizes the action transition even when the employed data is not used to calculate eigenvectors for principal component analysis. These results confirm that the proposed method is sufficient to follow dependent on the actions.