6-D Pose Estimation of Objects in Masked Image using Line Segment Detection

In this paper, for accurate grasping of objects, we propose a pose estimation algorithm that recognizes a three-dimensional object model within the mask area of the object detected by Mask R-CNN and estimates the 6D (degrees of freedom) pose of several instances. In industrial sites, it is challenging to determine the exact grasping position of small objects that lack characteristics when numerous parts are stacked. The proposed algorithm solves this problem by calculating the 6D pose of an object using the vertex detected from the contour of the object plane obtained from the image masked by instance segmentation. A line segmentation detection method in masked images based on low-cost RGB-D (Depth) cameras enables accurate pose estimation of objects for a robot gripper at high speed. The proposed method was verified through various experiments in an environment where rectangular objects were stacked.