image: Fig. 1. Optical-quasicrystal-coded single-shot stereo reconstruction.
Credit: Dayong Qiao, Yijie Shen et al.
Imagine trying to capture the three-dimensional shape of a complex object. Traditional high-precision 3D scanners might require illuminating the object with multiple light patterns and taking numerous photographs to piece together its shape. But what if a single flash could accomplish the task? Could a carefully designed single light pattern carry enough information to reveal the most subtle details of a moving object?
This is the challenge addressed by a research team led by Professor Dayong Qiao at Northwestern Polytechnical University and Assistant Professor Yijie Shen at Nanyang Technological University, together with collaborators at the University of Stuttgart. In a new study published in Light Dynography, the researchers demonstrate an approach that combines an unusual form of structured light, known as an optical quasicrystal, with artificial intelligence (AI) to reconstruct three-dimensional objects from a single synchronized stereo exposure.
The inspiration comes from a curious form of order found in nature. Ordinary crystals are built from patterns that repeat regularly, much like tiles covering a floor. Quasicrystals, however, follow a different rule: their structures are highly organized but never repeat in quite the same way. This unusual combination of order and nonperiodicity, first discovered in materials science, also offers intriguing possibilities for shaping light. When several light patterns are arranged to form an optical quasicrystal, they create a complex optical fingerprint that looks almost irregular but contains a carefully organized set of hidden directional features.
The researchers turned this optical fingerprint into a new language for 3D imaging. Instead of projecting multiple patterns one after another, they combined 12 differently oriented stripe patterns into a single quasicrystal design. Each direction carries a different piece of spatial information, yet the components remain distinguishable in the light pattern's frequency spectrum. As the projected light falls onto an object, its surface bends and distorts the pattern, encoding clues about the object's three-dimensional shape into the images captured by two cameras.
But packing more information into a single image is only half the story. The next challenge is learning how to read it. For this, the team developed an AI-powered reconstruction method called OQ-Stereo. Much like separating individual instruments from a complex musical performance, the algorithm disentangles the overlapping directional patterns and extracts complementary geometric clues. It then compares the two camera views and combines multiple depth estimates according to their reliability, transforming the encoded light into a detailed three-dimensional reconstruction.
The results show how much information a single flash can reveal. In experiments with planes, cylinders, and spheres, the system achieved sub-0.1 mm surface-fitting precision using just one projected pattern. It also reconstructed objects with intricate shapes, sharp edges, holes, and curved surfaces while retaining important geometric details. To test its performance on moving objects, the researchers turned to a rotating fan. Even as the blades changed position, individual captures produced recognizable three-dimensional reconstructions of the blades and central hub.
The significance goes beyond making a faster 3D scanner. Traditionally, higher measurement precision often comes at the cost of collecting more images, introducing a difficult trade-off between speed and accuracy. This study explores a different possibility: instead of asking a camera to capture more images, why not make each projected pattern carry more useful information? By combining the unusual physics of optical quasicrystals with the decoding capabilities of AI, the researchers have taken a step toward compact, high-precision 3D sensing systems that could one day help robots navigate complex environments, inspect moving industrial components, and capture detailed shapes in situations where every exposure counts.
Article Title
Optical Quasicrystal Coding for High-Precision Single-Shot 3D Reconstruction