High dynamic range fringe pattern acquisition based on deep neural network

Fringe projection profilometry has wide applications in three-dimensional measurement. When the fringes are projected onto the shiny surfaces with a large reflectivity, however, the saturation regions and dark regions are often generated on the surface due to the inconsistent reflection characteristics of the measured object. The saturation and dark regions make the modulation of captured fringe patterns degraded, and the accuracy of the three-dimensional measurement is significantly affected. To solve this problem, an improved network for high dynamic range fringe pattern is designed In this network, we introduce a low modulation region detection module based on U-net to obtain the saturation and dark regions in the fringe patterns. Then, the detailed fringe information in the low modulation regions can be predicted through a fringe enhancement module which takes advantage of the global distribution and detail intensity of the fringe patterns. In the experiment, a mass of simulated and actual low modulation fringe patterns are processed through the proposed network. The experimental results show that the designed network can effectively remove the noise of the captured fringe patterns and repair the low modulation regions captured from the shiny surfaces. Combined with phase calculation, the improved high dynamic range fringe patterns are used for three-dimensional data calculation. A standard metal gauge block with a height of 5 mm is measured and the root mean square error of the height is improved from 0.55 mm to 0.06 mm. This proves the proposed method can effectively improve the quality of fringe patterns projected on complex reflective surfaces.