Gaussian mixture model based invalid point removal for fringe projection profilometry

Abstract

In fringe projection profilometry, there inevitably exist many invalid points in the background and the occluded region that are not covered by the fringe pattern. The reconstruction result of invalid points is wrong, which seriously degrades the quality of the reconstructed point cloud and thus affects the measurement accuracy. Therefore, recognizing and removing invalid points is necessary. In this paper, an adaptive invalid point removal method based on Gaussian mixture model (GMM) is innovatively proposed for fringe projection profilometry (FPP). The proposed approach firstly calculates the modulation level using multistep phase-shifting method. Secondly, GMM is applied to model the density distribution of the modulation information and thus classify the pixels based on the modulation information. Thirdly, the classification results are further optimized using absolute phase gradient information and neighborhood information. Then the final classification results of the modulation intensity and the corresponding invalid point identification results can be obtained. A series of experimental results demonstrated that the method can accurately identify invalid points and improve the quality of point cloud reconstruction under different measurement scenarios.