Let’s Take a Walk on Superpixels Graphs: Deformable Linear Objects Segmentation and Model Estimation

Abstract

0000−0001−8203−9176] , Gianluca Palli 2[0000−0001−9457−4643] , and Luigi Di Stefano 1[0000−0001−6014−6421]

Auto-generated Wires Dataset for Semantic Segmentation with Domain-Independence

“…[25] developed a more sophisticated method that relies on Gaussian Mixture Models, but it requires the assumption of having a good color contrast between object and background (which has to be homogeneous). The state-of-the-art solution for DLOs detection is presented in [26]. This algorithm, called Ariadne, is based on biased random walks over the Region Adjacency Graph built on a super-pixel over-segmentation of the source image.…”

“…We evaluate and compare the outputs by means of the Dice coefficient Dice = 2 |Mp∩Mgt| |Mp|+|Mgt| , where M gt is the ground truth. Table I resumes the average Dice obtained in the test dataset by DeepLabV3+, HRNet and Ariadne [26], state-ofthe-art algorithm for DLO segmentation. Ariadne yields a bspline model for each wire which is here used as predicted mask M p .…”

Model-based Manipulation of Deformable Linear Objects by Multivariate Dynamic Splines

“…The algorithm is initialized by setting the control points at time step k = 0, let us call them q 0 , equal to the actual DLO configuration. As mentioned in the introduction, we assume at this stage that a vision system, such as the one reported in [14], can be used to detect the DLO and provide a spline-based representation of its shape (q 0 as it was named previously). Moreover, since we have no initial knowledge of the system input, i.e.…”

“…In this paper, a single-arm manipulation system equipped with a parallel gripper is assumed, and the investigation is based on the assumption that the robot can only grasp the DLO at a certain point and move point in another position on the same plane (the table plane). The DLO segmentation and model estimation from 2D images investigated in [14] will be exploited to provide the input model for the algorithm initialization after the execution of each manipulation primitive, i.e. grasping, repositioning of the cable at the grasping point and release.…”

New model-based manipulation technique for reshaping deformable linear objects

“…Moreover, the manipulation system is assumed to be able to detect the initial state of the DLO by using an appropriate vision system. For instance, the technique proposed in [7] may be used for providing the initial DLO state for starting the algorithm. Then, the manipulation technique will perform a series of grasping and releasing primitives for reshaping the DLO as desired.…”