Research on Bolt Pose Detection Technology Based on Deep Learning

Abstract

This paper focuses on the automatic operation of robots that automatically tighten bolts, and completes the work of bolt target detection and pose estimation. We utilize deep learning technology to detect bolts rapidly and accurately, achieving a precision rate of 97%. Next, based on image processing, the thread information of the bolts is extracted, and the posture information is estimated through clustering, in order to provide information support for robot operation planning. The experimental results show that our method achieves good results in both bolt target detection and bolt pose estimation, and provides an information basis for automated assembly operations.