Tracking Branched Deformable Linear Objects With Structure Preserved Registration by Branch-wise Probability Modification

Abstract

This paper focuses on tracking branched deformable linear objects (DLOs). It proposes a modification to the method of structure preserved registration (SPR) enhancing its capability for tracking DLOs with multiple branches. The modified method is applied to the practical problem of wire harness localization under occlusions. From a series of point clouds obtained by stereo vision, the positions of the individual branches of the wire harness are estimated. It is observed that SPR suffers from incorrect registration of branches in cases where single branches are completely occluded during manipulation, e.g., by a robotic manipulator. Therefore, we propose CAMP, a modified method of SPR, which relies on additional model information about the branched structure to avoid incorrect assignments by introducing branch-wise probability estimates in the underlying Gaussian Mixture Model (GMM). The performance of CAMP and its robustness against occlusion, is evaluated and compared to SPR in practical experiments with a wire harness.