Research On Fast And High-precision Pose Estimation Algorithm Of 3C Components By Using 3D Vision

3C is the collective name for computers,communications,and consumer electronics products.At present,the assembly process in the production of 3C products still requires a lot of labor.Robotic assembly of the 3C components can effectively reduce the labor cost of the enterprise and improve product quality.And the pose estimation of 3C components is a prerequisite for robotic assembly.For 3C components usually exhibit highly symmetrical planar structures,the traditional pose estimation algorithms based on 3D vision face many problems such as pose ambiguity,local optimal solution,and inefficiency.Therefore,this paper proposes a fast and high-precision pose estimation algorithm for 3C assembly.Our algorithm is divided into two phases: the global pose estimation and the local pose estimation.Among them,the former mainly provides an initial pose for the local pose estimation,and solves the ambiguous pose problem caused by 3C components with few features,high symmetry,and planar structures.In the latter stage,the improved iterative closest point(ICP)algorithm is applied to obtain a more accurate pose quickly.For the ambiguous poses problem of 3C components,the principal component analysis(PCA)and shape-based template matching are applied.Through PCA,multiple ambiguous poses can be obtained quickly.Then,the ambiguous poses can be eliminated through the shape matching of 2D images and the real initial pose is obtained.To solve the local optimal problem when ICP is applied to register the point cloud of 3C components,a method combining initial pose correction and application of translation bias is proposed.This method efficiently solves the local optimal problem of ICP.To boost the efficiency of the ICP in the local pose estimation stage,the relationship among the convergence accuracy,the calculation time of ICP and the initial pose accuracy,distance objective function,optimization algorithm,and the number of point clouds is analyzed.Finally,an improved point-to-plane ICP algorithm combining point cloud down-sampling and weighted linear least-squares optimization is proposed.This algorithm greatly improves the efficiency of the local registration stage.Through simulation and experiments,the reliability and effectiveness of the entire pose estimation scheme of 3C components are verified.The pose estimation scheme can eliminate ambiguity poses at 100%.For 3C components with planar structure,the total translation error is less than 0.20 mm,and the angle error is less than 0.24 °.In terms of efficiency,the entire scheme takes less than 1s,which meets the needs of the 3C assembly.