TPDNet: Texture-Guided Phase-to-DEPTH Networks to Repair Shadow-Induced Errors for Fringe Projection Profilometry

This paper proposes a phase-to-depth deep learning model to repair shadow-induced errors for fringe projection profilometry (FPP). The comprises two hourglass branches that extract information from texture images and phase maps fuses the by concatenation weights. input of proposed contains images, masks, unwrapped maps, ground truth is depth map CAD models. A loss function was chosen consider image details structural similarity. training data contain 1200 samples in verified virtual FPP system. After training, we conduct experiments on real-world scanning data, results support model’s effectiveness. mean absolute error root squared are 1.0279 mm 1.1898 validation dataset. In addition, analyze influence ambient light intensity performance. Low limits performance as cannot valid completely dark shadow regions images. contribution each branch network also investigated. Features texture-dominant leveraged guidance remedy errors. Information phase-dominant makes accurate predictions whole object. Our provides good reference repairing