Light-sectioning with many lines - how to settle the information deficit

The ideal optical 3D sensor displays a few key features: high accuracy, high lateral resolution, a large measurement volume and short acquisition time. Light-sectioning is one of the few principles that allows for high speed – a highly demanded property. However, if no context information is exploited, light sectioning with many lines displays an information deficit, leading to ambiguous correspondence. Last year we presented a novel method that partially settles this deficit by additional cameras. The method enables a significant increase of the number of projected lines, leading to a higher data density. However, the robustness is still depending on the object shape. Now we present a solution with a much better robustness, largely independent of the object. The idea is based on the following observation: Projected lines and their images that are back projected from the cameras generate periodic (Talbot-like) patterns in space. The overlap of these patterns encodes the correspondences. It turns out that there is a geometry which allows to find correct correspondences, depending only on the noise. We will demonstrate experiments and discuss the potentials and limits.