Fig. 1. Experimental device of weld surface quality detection based on line structured light

Fig. 2. Single-frame weld image after different processing algorithms. (a) Original image; (b) image after Gaussian filtering; (c) image after OTSU; (d) region of interest (ROI); (e) image after weld centerline extraction; (f) finally obtained weld centerline

Fig. 3. Schematic of feature points of weld surface

Fig. 4. Extraction of weld feature points. (a) Feature points of height; (b) welding toe points extracted by gradient method; (c) welding toe points extracted by K-means method

Fig. 5. Sample distribution on weld center line and K-means clustering effect. (a) Sample distribution; (b) original weld centerline; (c) initial clustering center; (d) final clustering center; (e) optimized result

Fig. 6. Comparison on weld width extracted by different methods. (a) Gradient method; (b) K-means clustering algorithm

Fig. 7. Three-dimensional weld model. (a) Point cloud model; (b) shape assembly; (c) three-dimensional illumination model

Fig. 8. Original images of three welds. (a) Flawless weld; (b) weld with undercut; (c) weld with porosity

Fig. 9. Feature extraction. (a)(d)(g) Intensity distributions collected by illumination model; (b)(e)(h) results after median filter; (c)(f)(i) results after binaryzation