Fig. 1. Binocular stereo vision imaging model

Fig. 2. Flow chart of stereo matching based on perspective transformation and SA-MLP network

Fig. 3. Perspective transformation principle

Fig. 4. Schematic of stereo matching based on cross laser

Fig. 5. Matching based on perspective transformation

Fig. 6. SA-MLP network structure

Fig. 7. SA-MLP algorithm flow chart

Fig. 8. Comparison of standard error before and after compensation of different models

Fig. 9. Effect before and after projection point compensation. (a) Before compensation; (b) after compensation

Fig. 10. Online detection of binocular vision

Fig. 11. Three-dimensional reconstruction and flatness results of circular hole under different matching methods. (a) Point cloud data generated with traditional matching method; (b) z-axis histogram of point cloud data generated with traditional matching method; (c) point cloud distribution map generated with traditional matching method; (d) point cloud data generated with proposed matching method; (e) z-axis histogram of point cloud data generated with proposed matching method; (f) point cloud distribution map generated with proposed matching method

Fig. 12. Extracted countersink hole edge contour

Fig. 13. Schematic of spatial cone model

Fig. 14. Experimental platform

Fig. 15. Overall framework of visual inspection system

Fig. 16. Monocular vision inspection experiment and inspection process. (a) Experimental setup; (b) filtering; (c) reflection suppression; (d) binary extraction; (e) sampling calculation

Fig. 17. Sample drawing of different types of countersink holes. (a) Hole type 1; (b) hole type 2; (c) hole type 3

Fig. 18. Error diagrams of monocular and binocular vision measurement for different pore types. (a) Hole type 1; (b) hole type 2; (c) hole type 3