Researchers have devised a quicker way to reconstruct images of 3D surfaces from fringe patterns. [Image: Ce Zhu, University of Electronic Science and Technology of China]
Researchers in China have reportedly developed a faster and more accurate method for creating detailed 3D maps of an object's surface (Optica, doi: 10.1364/OPTICA.531601). The new approach, which combines high frame rates with micrometer-level precision, could offer significant benefits for real-time scanning applications such as industrial inspection, medical imaging and machine vision.
Streamlining the process
Precise images of 3D surfaces are typically produced by projecting a series of sinusoidal light patterns onto the object, each one with a slightly different phase. The surface geometry modulates the phase of the light fringes, allowing a 3D map of the surface to be reconstructed by analyzing the phase changes in the reflected light patterns. However, several sets of images from multiple viewpoints are needed to achieve accurate surface measurements, which compromises the real-time performance of the technique.
In contrast, the new approach eliminates the data-processing step by exploiting other information contained in the phase data. In particular, the researchers show that the variation in depth across the surface is directly related to the phase gradient, providing a crucial reference point that enables the surface profile to be reconstructed from a single set of phase-shifted patterns.
Fringe photometric stereo
The researchers tested their method, which they call fringe photometric stereo (FPS), by creating 3D surface images of various objects, such as a human hand, a cloth toy and a collection of geometric structures made from plaster. Their results, obtained using just a standard projector and camera, show that FPS can accurately recover the 3D surface profile of continuous surfaces without any additional processing. The depth of isolated objects can also be measured, but in this case an approximate reference depth must also be obtained by projecting a random speckle pattern onto the object.
The researchers also compared the noise performance of FPS with the conventional profiling approach. When taking measurements of a standard ceramic plate, they show that FPS can suppress noise by about 50% while also reducing the number of patterns needed to reconstruct the surface profile from 24 to eight. Further tests with a standard sphere confirmed that FPS can mitigate for noise and maintain a consistent depth resolution of around 50 µm.
One remaining problem is that FPS can struggle to reconstruct complex 3D scenes with abrupt changes in depth. "We aim to address this issue by accurately identifying depth jumping areas and assigning them lower weights, which may be key to extending FPS to broader applications," comments team leader Ce Zhu from the University of Electronic Science and Technology of China.