Target Round Bar Detection and Two-stage Calibration for Automatic Label Attachment System

In this paper, we propose a target round bar detection method and two-stage camera calibration method for an automatic label attachment system. The proposed method allows a collaborative robot to automatically and accurately attach labels to round bars. In order to detect target round bars, a round bar detection network is trained by using a round bar image with annotations, and the learned round bar detection network detects round bars in captured images. We propose a hierarchical clustering technique to remove the other round bars except the target round bars. It is important not only to accurately detect the round bars in images, but also to precisely convert the location information in the image into coordinates in the real world. A twostage camera calibration method is proposed to perform precise coordinate transformation. We test and evaluate the proposed method using test images obtained from real sites. The experimental results show that the proposed target round bar detection method provided high detection accuracy and accurate round bar center estimation ability while detecting only the target round bars. They further show that the proposed camera calibration method gives precise coordinate transformation performance through more than hundreds of automatically generated coordinate pairs.