By Andrew Neill
Computer Vision Final Project
EGGN 512
Outline
Topic
 Motivation
 Related Literature
 Experiments Conducted
 Results
 Discussion and Conclusions
 References
 Code Appendix

Topic

Post-Process Weld Bead Stereo
Imaging
 After welding process, what is the profile of
the weld
 Unique aspects of stereo imaging of welds
Motivation
Post-process evaluation
and qualification
 In-Process evaluation
 Process control

 Refinement of processes
based on weld profile
Related Literature – Stereo
Vision

Countless algorithms for stereo
correspondence
 Scene Reconstruction
 Occlusions, processing time, multiple inputs

High detail stereo imaging
Related Literature – Welding
Stereo Vision

Pre-process weld groove identification
 Chen, Chen Qiu and Li – Weld seam
identification

In-Situ weld pool modeling
 Chris Mnich – High-speed stereo vision of
a weld pool
○ Created 3D model of weld pool over time

Post-process weld qualification
 Seyffarth and Gaede – laser profile
scanning
○ Accurate and higher precision
○ Computationally expensive
Experiments
Tried my own simple algorithms for
stereo imaging
 Unique points to this problem

 Difficulties arise with lighting, specular and
anisotropic reflection, and identical features
which cause a significant amount of noise
 Advantages are present in the large amount
of texture and smooth nature of the material
Experiments

Control experiment using a bolt
 Optically similar metal
 High amount of texture
 Ridges of larger dimensions to extract

Lighting
 Flash, no flash, direct lighting, indirect lighting

Averaging
 Using different template sizes

Smoothing
 Gaussian smoothing on disparity map
Results – Stereo Imaging

Camera directly over
sliding table
Results – Bolt
Original
Disparity maps
Results – Lighting
Camera Flash
Full Direct Background
Lighting
Half Direct
Background Lighting
Results – Lighting

Indirect background
lighting produces the
best results
 Low specular
reflections
 Most consistent
shadows
Results – Stereo Correspondence
Undistorted original image
 Cropping and shifting image
 Used cross correlation algorithm
 Employed search window to speed calculations

Results – Smoothing

Unsmoothed Disparity
Map
 Points are sporadic and
hard to see
 3D graph is too noisy to be
useful

Smoothed Disparity Map
 Features become apparent
 Noise is significantly
reduced
 Mesh graph is smoother
and easier to comprehend
Results – Averaging

Averaged 3 template sizes
11x11
Averaged Together
17x17
21x21
Discussion

Bolt
 Found and easily identified
 Disparity map matches shape
 Algorithm worked very well to find the bolt

Lighting
 Difficulties arose with reflections but were solved with indirect
lighting
 Lighting was very important to ensure disparity matching

Stereo Imaging
 Photographing method was effective
 Required shifting the image for effective

Averaging
 Yielded no significant advantage

Smoothing
 Significant improvement for visualization
Conclusions
Significant amount of work being done
for weld pool analysis and seam tracking
 Initial weld stereo results were very
poor, lighting was instrumental in helping
to correct this
 Final results were still poor, but a human
eye can still see the weld location
 This small research project has barely
scraped the potential for this application

References





Chen, S. B., X. Z. Chen, T. Qiu, and J. Q. Li. "Acquisition of
Weld Seam Dimensional Position Information for Arc Welding
Robot Based on Vision Computing." 43.1 (2005): 77-79.
Mnich, C., Al-Bayat, F., Debrunner, C., Steele, J., Vincent, T. "In
Situ Weld Pool Measurement using Stereovision." ASME
Symposium on Flexible Automation (2004).
Mnich, Chris M. Development of a Synchronozed, High-Speed,
Stereovision System for In Situ Weld Pool Measurement.
Masters Thesis. Colorado School of Mines. Golden, Colorado,
2004.
Seyffarth, Peter, and Rainer Gaede. "Image Processing for
Automated Robotic Welding." Ingenieurtechnik Und
Maschinenbau GmbH Rostock (n.d.).
Zhang, Y. "Characterization and Real-time Measurement of
Geometrical Appearance of the Weld Pool." International
Journal of Machine Tools and Manufacture 36.7 (1996): 799816.