BP neural network applied to 3D object measurement based on fringe pattern projection

In the fringe projection profilometry, the traditional triangle method, such as Fourier-transform profilometry (FTP), is difficult to recover the stepped shape object from the deformed fringe pattern. In order to solve this problem, the neural network is introduced to deal with this kind of fringe patterns and gain the three-dimensional (3D) information of the measured object. By training the network, the relationship between the deformed fringe pattern and the height of the object can be obtained, and thereby the height of the object can be obtained. Furthermore, the object can be reconstructed perfectly without knowing the optical parameters of the experiment system. An obvious merit of this network method is that it can recover the 3D object in a short time and only need one deformed fringe pattern. Computer simulations and experiment validate the feasibility of the method.