Research On Sparse Monocular Viewpoints Based Pose Estimation Technology Of Texture-less Parts

Under the background of "robot instead of human",robot vision intelligence is of great significance to the industrial upgrading of manufacturing industry,and the industrial robots with the visual sensors and the intelligent algorithms are increasingly valued in industrial production.The pose estimation of parts with vision technology and the grasping and assembling with robots is the core of robot vision intelligence.For the texture-less metal parts commonly used in industrial scenes,there is little information available in the image.Most of the current techniques use the lowest features of the edge pixels in the image to calculate the pose,which cannot estimate the pose fast and accurately.In this paper,a new pose estimation method frame for texture-less metal parts is proposed,which can achieve high-precision and robust pose estimation with sparse templates.First,the geometric features of the part are used to match the actual image and the template.Then the special position points on the matched geometric features are used to calculate the accurate pose.Specifically,we select the relationship among the straight line contours as the high-level geometric features describing the metal part.And the endpoints of the matched straight line contours between the actual image and the template are used to calculate the pose accurately.Further,based on the existing Line Segment Detector(LSD)and Bunch Of Lines Descriptor(BOLD),this paper proposed the Complete Line Segment Detector(CLSD)to detect straight contours and the Grayscale-Invert-Invariance BOLD(GIIBOLD)to describe the straight contours,which solved the shortage of LSD and BOLD.Finally the EPnP algorithm is used to calculate the accuracy pose to implement the above method frame.For the sparse monocular viewpoints based pose estimation of texture-less parts,the monocular sparse viewpoints templates generation,the complete line contour feature extraction,the description based on the bunch of lines features and the pose estimation based on the sparse templates are researched in this paper.The software and hardware platform of the pose estimation and grasping system of the texture-less metal parts is established.And the pose estimation and grasping of the texture-less metal part is completed with this platform successfully.