Robot out-of-order target sorting method based on 3D visual clustering and matching

Abstract

The invention discloses a robot out-of-order target sorting method based on 3D visual clustering and matching, and the method comprises the steps: filtering a sorting scene point cloud plane of collected sorting scene point clouds, carrying out segmentation of the point cloud plane of a stacked sorting scene, and obtaining a plurality of types of sorting scene point cloud clusters; querying a three-dimensional point neighborhood to obtain a stacking state of each sorting scene point cloud cluster; estimating the normal of each point in the sorting scene point cloud, and extracting the edge ofeach sorting scene point cloud cluster and the edge of the target model; generating different edge candidate matching sets, acquiring an initial pose through Super4PCS coarse matching, and conductingsorting recognition and pose estimation through ICP fine matching. By means of three-dimensional perception of the objects, recognition, classification, matching and positioning of the multi-target objects with the characteristics of stacking, shielding, disorder and the like are achieved, autonomous recognition of an industrial robot and planning of grabbing actions are facilitated, and thereforethe grabbing efficiency and accuracy degree of carrying and sorting operation are improved.

Description

technical field [0001] The invention relates to a method for sorting and recognizing robots out of order objects, in particular to a method for sorting out of order objects by robots based on 3D vision clustering and matching. Background technique [0002] With the widespread application of industrial robots in the fields of intelligent sorting, assembly and manufacturing, the production mode of enterprises has gradually shifted from the traditional human-led to robot-led. The traditional sorting method mainly relies on workers for multi-variety and small-batch workpieces. This method requires high employment costs. With the increase of duration and work intensity, the sorting accuracy will be greatly affected, and due to the labor time limit, The work efficiency is not high. With the transformation of the manufacturing industry to automation, intelligence and informatization, the cost of using machine vision is gradually reduced, and the robot sorting system equipped with ...

AI Technical Summary

Problems solved by technology

[0004] However, since the effective feature descriptors of 3D point clouds, such as FPFH, SHOT, VFH, etc., have certain limitations in the recognition range and accuracy, they cannot well adapt to the production environment with multiple scene changes and workpieces with multiple varieties and poses. The lack of universality and universality restricts the wide application of 3D vision in practice

3D point cloud recognition mainly has the following problems: ①The visual device usually only acquires limited data from a single perspective, and there are point cloud incomplete phenomena caused by workpiece stacking, environmental constraints, sensor performance, etc.; ②The normal consistency of some regular workpieces is poor. High, most point cloud descriptors based on normals have little effect; ③ workpieces with similar shapes and structures have similar characteristics, and it is difficult to accurately identify them

Images