Multichannel autostereoscopic measurement system for micro-structured surfaces based on multi-scale depth fusion

Abstract

Autostereoscopic technology has been acknowledged to realize precise metrology of micro-structured surfaces. While autostereoscopic performs well, the reliance on a single light field modality constrains the performance of the system. To address this issue, a multichannel autostereoscopic measurement (MAM) system has been developed to provide richer data for depth estimation. The whole system includes a 3D optical channel to capture elemental images (EIs) from various viewpoints, in conjunction with a 2D channel to obtain high-resolution (HR) images. This system employs data fusion techniques to compensate for data deficiencies and enhance accuracy. In the 2D channel, a deep learning network called UniDepth is used to estimate the 3D geometry of objects based on HR images. Combining multi-scale depth information from different optical channels via the pyramid representation allows for more precise 3D reconstruction. The experimental results demonstrate that the proposed multichannel measurement system improves the quality and robustness of 3D reconstruction.