Visual Recognition Method for Deformable Wires in Aircrafts Assembly based on Sequential Segmentation and Probabilisitic Estimation

Manipulation object recognition is necessary for robotic manipulation planning and control. However, wire harnesses to be automatically assembled in aircrafts have small radial scale and complex structure which make them hard to be recognized. A method in order to accomplish multi-branch wire harness objects recognition is presented, which involves segmentation and estimation. Segmentation is performed in a sequential approach based on Cartesian distance, color similarity and bending continuity, among which the bending character considering rigidity diminishing thesis provides the segmentation result with physical interpretability. Gaussian Mixture Model is employed to making an estimation to complete the missing wire parts resulted from occlusion and omission on the segmented results. Effects of the method have been verified on a point cloud dataset with over 200 aviation wire objects. It has been demonstrated that aviation wire harnesses with multi-branches and occlusion can be distinguished and completed with a success rate over 96%.