Three-dimensional shape measurement based on projector defocusing and deep learning

Abstract

High-speed real-time three-dimensional(3D) measurement is of great significance for the measurement of high-speed dynamic process. However, the projector needs a long time integration to project different gray levels, and needs a complicated gamma correction process, which greatly limits its measurement speed and accuracy. The appearance of defocus fringe projection technology provides a new perspective for high-speed dynamic 3D measurement. With the research on the feasibility of the combination of deep learning and traditional digital fringe projection 3D measurement technology, the development of fringe projection technology is also breeds new breakthroughs. In this paper, we combine the defocus fringe projection technology with deep learning, and propose a 3D shape measurement based on projector defocusing and deep learning. Experiments show that by collecting sufficient and comprehensive training data and after many reasonable trainings, this method can effectively eliminate the ripple error in the traditional three-step phase shift method and the measurement accuracy is comparable to that of the twelve-step phase-shifting method.