A high-precision and noise-robust subpixel center extraction method of laser stripe based on multi-scale anisotropic Gaussian kernel

Accurate center extraction of laser stripe is essential for active 3D visual reconstruction with high precision. However, the existing methods poorly balance accuracy and noise robustness in the actual industrial measurement. This paper proposes a new high-precision and noise-robust subpixel laser stripe center extraction method based on multi-scale anisotropic Gaussian kernel. Firstly, we modeled the laser strip via a Gaussian function, and then determined the second order maximum response and its direction through the second order anisotropic Gaussian kernel. Finally, the laser strip subpixel centers are calculated through Taylor expansion. The proposed method is evaluated by the accuracy and robustness to direction, noise and width. The experimental results showed that the extraction accuracy is less than 0.2 pixel at high noise levels and the F-Measure is not less than 0.96 with the increase of noise. Thus, the proposed method achieves high-precision and noise-robust extraction for laser stripe subpixel center.