Sparse Template-Based 6-D Pose Estimation of Metal Parts Using a Monocular Camera

The six-dimensional (6-D) pose estimation of smooth metal parts is a common and important task in intelligent manufacturing. Computer-aided design (CAD)-based monocular vision methods offer more advantages than those offered by other methods. However, they are subject to several drawbacks such as high complexity, low robustness, and unsatisfactory accuracy, which hinder their application in industry. In this paper, a new approach with corresponding practical algorithms is proposed to solve these problems. The proposed approach uses high-level geometric features and the correlation of straight contours, to represent the part. Moreover, it exploits the matched special location points on the geometric features, which are the endpoints of the straight contours, to accurately estimate the 6-D pose. Practical algorithms based on the modification of the existing line-feature descriptors are proposed to implement the approach. The experimental results revealed that the proposed approach and algorithms can achieve higher accuracy and robustness with fewer templates.