Fig. 1. Principle of laser triangulation based on dual-reflectors. (a) Overall optical path structure; (b) specific optical component position parameters at imaging end

Fig. 2. Measurement sensitivity analysis. (a) Relationship between sensitivity and reflection angles θ₁ and θ₂; (b) relationship between sensitivity and image distance

Fig. 3. Principle of self-supervised denoising network ZS-N2N

Fig. 4. Experimental verification platform. (a) Actual optical path of laser triangulation system based on dual-reflection; (b) extract ROI from grayscale image captured by CMOS, and analyze deviation of spot in x and y directions

Fig. 5. Three dimensional distribution maps of noise corresponding to different noise coefficients. (a) Noise coefficient of 1; (b) noise coefficient of 2; (c) noise coefficient of 5; (d) noise coefficient of 10

Fig. 6. Denoising performance of different methods. (a) PSNRs corresponding to different noise levels; (b) RMSEs corresponding to different noise levels

Fig. 7. RMSE of localization results in X and Y directions using different methods under different noise levels

Fig. 8. Distribution of deviation between centroid position and true position after using different methods at actual noise levels; (a) ZS-N2N method; (b) median filter; (c) Gaussian filter

Fig. 9. Running time using different methods under different noise levels

Fig. 10. Analysis of repeatability of systems with different working distances. (a)‒(e) Deviation curves of centroid position of spot in X and Y directions when working distances are 1750, 2000, 2250, and 2500 mm, respectively

Fig. 11. Repeatability of systems after denoising spot using ZS-N2N method, median filtering, and Gaussian filtering

Fig. 12. Five kinds of different surface roughness objects

Fig. 13. Repeatability of systems facing different object surfaces at different working distances. (a)‒(d) Repeatability of systems facing five different object surfaces at working distances of 1750, 2000, 2250, and 2500 mm, respectively