Accurate defocusing fringe projection profilometry in a large depth-of-field

Defocusing fringe projection profilometry (DFPP) has been one of the most popular 3-D measurement techniques. The measurement error caused by the low-contrast patterns becomes non-ignorable in large depth-of-field (DoF) DFPP. Traditional methods sacrifice the measurement speed, 3-D details or ignore the influence of the projector defocusing, which limit the performance of extending the system DoF. In this paper, a deep learning-based fringe-enhancing method (DFEM) is proposed, which transforms three patterns with different phase shifts captured at a fixed focal length into the desired phase. DFEM divides multiple sub-DoFs for reducing the difficulty of pattern transformation in the training. In the testing, the geometric constraint is introduced for determining the sub-DoF in which the object is located. DFEM can achieve accurate 3-D reconstruction in a DoF large to 1420 mm, which improve the DoF of DFPP up to 4.7 times of the traditional one. The provided experiments demonstrate the accurateness of the resulted large-DoF DFPP.

Automated summary

In this paper, a deep learning-based fringe-enhancing method (DFEM) is proposed to improve the accuracy of 3-D reconstruction in large depth-of-field DFPP. The DFEM divides multiple sub-DoFs for pattern transformation in training and introduces geometric constraint for determining object location in testing. The experiments demonstrate the improved performance of DFPP with a larger depth-of-field.