mBEST: Realtime Deformable Linear Object Detection Through Minimal Bending Energy Skeleton Pixel Traversals

Robotic manipulation of deformable materials is a challenging task that often requires realtime visual feedback. This especially true for linear objects (DLOs) or “rods”, whose slender and flexible structures make proper tracking detection nontrivial. To address this challenge, we present mBEST , robust algorithm the DLOs capable producing an ordered pixel sequence each DLO's centerline along with segmentation masks. Our obtains binary mask then thins it to produce skeleton representation. After refining ensure topological correctness, pixels are traversed generate paths unique DLO. At core our algorithm, postulate intersections can be robustly handled by choosing combination minimizes cumulative bending energy DLO(s). We show simple intuitive formulation outperforms state-of-the-art methods detecting large numbers sporadic crossings ranging from curvatures high variance nearly-parallel configurations. Furthermore, method achieves significant performance improvement approximately 50% faster runtime better scaling over state art.