Abstract 

II 

Abstract 

In recent years, industrial manufacturing technology development, welding metal 

processing as the important link, its requirements also more and more high, in order to reduce 

artificial factors on the influence of the welding quality, simplify the production process, 

improve the production efficiency, we must realize the welding automation, intelligent, this 

also is the necessary direction of welding industry. 

Weld automatic tracking is the key to realize automation of welding, laser visual guide 

intelligent welding robot include weld image recognition and robot motion control technology, 

can effective solve the weld automatic tracking problem. 

In order to realize the laser visual guide seam tracking purpose, with six axis DENSO 

industrial mechanical arm as the main body, a visual guide seam tracking platform, constructs 

the mathematical model of structure light guide weld robot, including camera linear model, 

nonlinear model of the camera, the camera external reference model, linear structured light 

measurement model and robot visual hand-eye model. 

The image processing part of laser visual guide seam tracking system mainly including 

target recognition, matching and weld feature extraction. The checkerboard target which for 

camera calibration, use Harris corner detection, and based on Hough translate and minimum 

between class variance to make image coordinates and the world coordinate match. Design a 

simple target, simplify the center point extraction and feature point matching algorithm, 

improve the matching algorithm robustness, reduce the operation time. Introduce the image 

preprocessing method for weld image in hardware and software , analysis and contrast the 

weld image binarization algorithm, analysis different kinds of weld image, and to different 

weld using different identification method and test. 

Six shaft mechanical arm whole calibration method, firstly analyzes the whole model, 

use the Zhang calibration method for camera calibration, use LM algorithm to the parameters 

nonlinear optimization. And then improve the algorithm plane equation use dual relationship , 

avoid error data influence; In the use of quaternions solving equation, and use the quaternion 

eliminate error data; Finally, according to the uncertainty of the welding parts, through the 

simple point contact method, correction the welding parts in the coordinates. The actual test 

shows that this method is robust and can get stable matching results, calibration of the camera, 

get hand-eye matrix and line laser plane equation stability, plane tracking can achieve ±0.3 

mm, 3 D tracking can achieve ± 1.2 mm , meet the industrial field application requirements. 

With calibration experiments, the tracking system of the key technology in the model: 

introduce transformation between the each coordinate, and using delay tracking strategy   

against the strong interference, so as to improve the stability of the seam tracking system, And 

in VS2010 environment write experimental procedures, through the test, the system can be 

stable, accurate weld automatic tracking，racking can achieve ± 1.4 mm. 

Keywords: Weld extraction, laser vision sensor, hand-eye calibration, seam tracking