Depth extraction for shinny objects from a single-frame fringe image

Fringe projection profilometry has been a mature technique for 3D measurement. However, it may fail if the measured surface shows saturations due to specular reflection. Numerous existing methodologies address this issue by increasing computational or hardware costs, while a single-frame strategy is lacking. To this end, this paper presents a single-frame fringe depth measurement for shiny objects. The cause of specular reflection is analyzed first, based on the Lambert and Blinn-Phong models, to disclose some clues embedded in the cause of oversaturation. With this, a ResUNet is constructed to extract the normal map from a fringe image and repair the missed fringes in the saturation field separately. The complemented fringe image and the normal map are subsequently fed into the designed DCAM-MFUNet to predict the depth. Extensive experiments demonstrate the great effect of our method.