Abstract: To solve the problems of scattered point interference and data holes in 3D target reconstruction, and to improve the 3D visual reconstruction effect, a 3D visual reconstruction algorithm based on polarization multispectral fusion was proposed. A binocular laser scanning and polarization multispectral imaging system was built, and a fusion algorithm was used to filter the point cloud by considering the characteristic region of multispectral mapping as the two-dimensional boundary of the target 3D point cloud. The precision of Gaussian and extremum sampling was tested experimentally. The mean deviation between the mapping position and the actual position was 0.59 mm and 0.93 mm, respectively. The average noise intensity in the background area was reduced from 49.5 to 13.4 after the superposition of four polarizers for noise reduction. Testing the target features with two different local curvatures determined that after optimization, over 80% of the test points had an error better than 3.05 μm, with an average deviation of 1.49 μm. Moreover, the 3D visual reconstruction effect of the target was improved.