Research on Visual Servoing control of metal objects in complex and variable lighting environment

The problem of recognizing metal parts under complex and variable lighting conditions in conventional visual servoing is addressed. In this paper, we combine the deep learning semantic segmentation network with the visual servoing algorithm to significantly reduce the dependence of target recognition on stable illumination. A semantic segmentation model is used to extract the pixel regions of the target features, and then these features are input into the visual servoing controller to move the robotic arm to the area around the target.