A method for high-precision grasping of symmetrical workpieces using low-precision depth cameras

Abstract

The invention relates to a method of grabbing a shape symmetrical workpiece with a high precision by using a low precision depth camera. The method comprises the following steps: 1) measuring a position of a conveying workpiece tray in a grabbing area of a production line in a grabbing coordinate system O-XYZ; 2) carrying out hand-eye calibration on a grabbing system of a mechanical arm to obtaina homogeneous coordinate transfer matrix T1 between a camera coordinate system and the grabbing coordinate system; 3) generating a 3D workpiece model point cloud and marking the grabbing position; 4)constructing a workpiece detection network model based on deep learning and training the model; 5) carrying out 2D image target detection according to the trained workpiece detection network model; 6)integrating an image target detection result and tray position information, and cutting the 3D workpiece model point cloud to obtain a workpiece observation point cloud; and 7) estimating a grabbinggesture according to the cut workpiece observation point cloud. Compared with the prior art, visual guidance can be carried out by using the low precision depth camera to achieve the grabbing precision with visual guidance by a high precision depth camera.

Description

technical field [0001] The invention relates to the field of mechanical arm control, in particular to a method for grasping a symmetrically shaped workpiece with high precision using a low-precision depth camera. Background technique [0002] Traditional industrial robot grasping usually adopts teaching programming mode. This method is suitable for stable working conditions, and the position and posture of industrial parts (hereinafter referred to as workpieces) are fixed. If the position and posture of workpieces are not fixed, it will A fetch failure occurred. There are various ways to solve this problem, and the most direct way is to use machine vision similar to human eyes to guide industrial robots to work. [0003] In recent years, with the widespread use of RGB-D depth cameras, the robot's three-dimensional perception of the environment has been continuously enhanced, and multi-modal data of the target can be easily collected. But in the actual environment, the imag...

AI Technical Summary

Problems solved by technology

[0004] At present, the target detection method based on deep learning is the most advanced method, but although this method can detect the target correctly, it cannot guarantee that the rectangular frame used for positioning completely contains the target, which will lead to loss when using the rectangular frame to segment the point cloud Part of the target point cloud

For dense point clouds acquired with high-precision depth cameras, the loss of a small part of the point cloud has little effect on subsequent target model fitting and pose estimation, but high-precision depth cameras are expensive, which means high production costs for manufacturers cost

Low-precision depth cameras are cheap, but the obtained point cloud is sparse, and the loss of a small amount of point cloud will seriously reduce the accuracy of the capture pose estimation, resulting in capture failure

Images