Mechanical parts picking through geometric properties determination using deep learning. (2nd February 2022)

• Record Type: Journal Article
• Authors: Lee, YJ; Lee, SH; Kim, DH
• Is Part Of: International journal of advanced robotic systems. Volume 19:Number 1(2022)
• Publication Date: 2022-02-02
• Subjects: Parts picking robot -- deep learning -- robot vision -- YOLO; Robotics -- Periodicals; Robotics; Periodicals; 629.892
• DOI: 10.1177/17298814221074532
• Languages: English
• ISSNs: 1729-8806

Abstract

In this study, a system for automatically picking mechanical parts required in the industrial automation field was proposed. In particular, using deep learning, bolts and nuts were recognized and geometric information of these parts was extracted. By applying YOLOv3 specialized in high recognition rate and fast processing speed, the recognition of target object, location, and postural information were obtained. The geometric information for the bolt can be obtained by creating two bounding boxes and calculating the orientation vector formed by these center values of two bounding boxes after successfully detecting two individual bounding boxes. Moreover, to obtain more precise geometric information on bolts and nuts, image distortion compensation on the detected object was done after detecting the center value of the bolt and nut through YOLOv3. Based on this result, it was proven that an automatic picking of the mechanical parts using a five-axis robot was successfully implemented.