Object Recognition And Grabbing Based On Vision

Automatic capture is one of the basic tasks of industrial robots in the production operation,the general implementation using the traditional off-line teaching trajectory in advance to set the robot,repeatability of the structured degree and grasp task depends heavily on the work environment of the robot,the dynamic process is not suitable for modern flexible manufacturing of products with different specifications.This paper is based on the modern intelligent manufacturing of robot automatic capture task requirements,for lack of robot job target perception information,focusing on the workpiece positioning and visual capture method based on robot to improve the unstructured task adapt ability,automatic detection and meet the operation requirement of grasping different objects.This paper first analyzes the typical structure and task applicability of robot vision guidance system,and builds a hardware experiment system of robot self-grasping based on Eye-To-Hand structure by using Kinova robot and Kinect.Then studied the Tsai-Lenz hand eye calibration algorithm,the calibration of robot grasping hardware system is build up in this topic,and,by the way of visualization shows the hand e ye calibration results,the average position error of hand eye calibration for 4.398 mm,the maximum position error is 8.048 mm,the average error angle is 1.322 degrees,the maximum angle error 2.227 degrees.Secondly,in order to improve the perceived qual ity of the Kinect scene,the empty noise existing in the shooting depth map Kinect,this paper proposes a depth map guide repair algorithm based on sampling,image has lower mean square error compared with MC-UE algorithm,the Kinect depth map obtained mor e realistic scenes.Then,the Drost-PPF algorithm research and realization method,realizes dynamic target recognition and pose estimation,and solve the problem of accuracy loss leads to the Drost-PPF algorithm in the discretization process,introduces th e ICP algorithm of Drost-PPF algorithm results postprocessing algorithm,the pose estimation results were corrected by ICP.After the ICP correction,the cumulative error between the model and the scene target correlation is only 6mm.Finally,a large number of grasping experiments were carried out on the robot self grasping platform built in this paper,which verified the visual grasping system and related algorithm built in this paper,which is effective for the self grasping tasks of centimeter level multi specification targets.The research work of this topic is of theoretical significance for enriching the related technologies of robot hand eye coordination and autonomous grasping,and has certain engineering application value.