System nonlinearity correction based on a multi-output support vector regression machine

Abstract

In a fringe projection profilometry system, the phase error introduced by the projector's gamma distortion is the main source of errors. To overcome this problem, we present a phase compensation scheme for multi-dimensional harmonic coefficient prediction based on a multi-output support vector regression machine(M-SVR), The scheme first constructs a significant characteristic relationship between phase probability density function (PDF) and phase multi-harmonic coefficients, creates simulation data without a priori knowledge, constructs a data set with a certain sample size, and then trains the M-SVR model. The trained M-SVR model is used to capture the potential features of the experimental distorted phase and output the multi-dimensional harmonic parameters with nonlinear relationships, followed by error compensation of the distorted phase using an immobile point iteration algorithm for the purpose of correcting the system nonlinearity. We demonstrate the validity and stability of the model through simulation and experimental trials. Most importantly, the preprocessed M-SVR model also has the potential to participate in error correction of other measurement experiments with reasonable sample and hyperparameter settings, which greatly saves the time and cost of multiple experiments.