Deep learning based measurement accuracy improvement of high dynamic range objects in fringe projection profilometry

One of the key factors affecting the accuracy of three-dimensional (3D) measurement in fringe projection profilometry (FPP) is the phase retrieve accuracy. In the 3D measurement of high dynamic range (HDR) objects, fringe saturation and/or low contrast are difficult to avoid. A greater number of fringe images are needed for 3D measurement of HDR objects by traditional methods, which is unfavorable for the measurement of moving objects. In this paper, what we believe to be a new method to solve the phase demodulation problem of HDR objects using deep learning is proposed. In this method, a "many-to-one" mapping relationship is established using an improved UNet deep neural network. In addition, in order to obtain more saturated fringe information, π-shifted binary fringes were also used. This allows us to retrieve the wrapped phase of HDR objects quickly and accurately. Experimental results demonstrate the effectiveness and reliability of the proposed method.