Three-dimensional (3D) imaging is crucial for various applications, including light detection, autonomous vehicles, and gesture recognition. While structured light techniques offer excellent performance, traditional projector devices are bulky due to refractive lenses and other components. Diffractive optical elements (DOEs) suffer from limited field of view (FOV). Metasurfaces, offering flexible control of optical wavefronts, are promising for compact 3D imaging devices. Existing metasurface-based approaches have limitations, such as limited FOV, depth of field, and image resolution. This research aims to address these limitations by proposing a novel single-shot 3D imaging system using a single-layer metasurface to project a coded point cloud and a corresponding reconstruction algorithm based on triangulation. The goal is to develop a compact, high-accuracy 3D imaging platform suitable for various practical applications.

The paper reviews existing 3D imaging techniques, highlighting the advantages and disadvantages of structured light methods and diffractive optical elements. It discusses the use of metasurfaces in holography, conformal optics, and beam shaping, emphasizing their potential for miniaturization and enhanced functionality in 3D imaging. The authors point out limitations in existing metasurface-based 3D imaging techniques, such as limited FOV, depth of field, and resolution, motivating their proposed approach.

The proposed method employs a single-layer metasurface to project a designed pattern in Fourier space. The pattern is composed of random point clouds, with local patterns uniquely identifiable across the projection plane. The 3D reconstruction leverages the triangulation principle, combining different metasurface and camera positions. To address speckle noise, a calibration and reconstruction operation is introduced using a reference plane and auxiliary planes. A matching algorithm, incorporating feature-based initial matching and area-based fine matching, is developed to establish accurate correspondences between captured and reference images. A multiresolution search strategy further enhances matching accuracy and efficiency by employing a pyramid strategy, starting with low-resolution images to establish coarse depth maps and refining the result with higher-resolution data. The metasurface is designed using M-array coding for local pattern uniqueness. The Gerchberg-Saxton (GS) algorithm and rigorous coupled wave analysis (RCWA) are employed for phase hologram calculation and nanopillar optimization. Amorphous silicon nanopillars are fabricated using electron beam lithography and reactive ion etching. The zero-normalised sum of squared difference (ZNSSD) is used to assess the similarity between images at different depths.

The research demonstrates a depth accuracy of less than 0.24 mm at a measurement distance of 300 mm. The proposed matching algorithm effectively handles speckle noise, enabling accurate 3D reconstruction. The multiresolution search strategy enhances both accuracy and efficiency. The system successfully performs 3D imaging on various scenes, including deformable cardboard and hand gestures, even with objects exhibiting low texture or varying reflectivity. The spatial resolution is shown to improve with an increased number of projected points, indicating scalability. The peak valley (PV) value is less than 0.24 mm, and the root mean square (RMS) value is 4.4 × 10⁻⁴ mm, showing high accuracy and planeness. The authors demonstrate 3D shape reconstruction of continuous and low-texture surfaces (deformable cardboard) as well as discontinuous objects with varying reflectivity (hand gestures), showcasing robustness and adaptability.

The results demonstrate the feasibility and effectiveness of the proposed single-shot 3D imaging system. The high accuracy and robustness of the system, despite challenges such as speckle noise and varying surface reflectivity, suggest its potential for real-world applications. The ability to image objects with low texture is a significant advantage over passive imaging techniques. The integration of nanophotonics and computer vision offers a path towards compact and cost-effective 3D imaging devices. The modular nature of the system allows for flexibility in FOV and dot density.

This study successfully demonstrates a novel single-shot 3D imaging technique utilizing a metasurface. The combination of a unique coded projection pattern, a sophisticated matching algorithm, and a multiresolution search strategy results in a high-accuracy, robust system. The experimental results with deformable materials and hand gestures validate its efficacy for various scenarios. Future research could focus on further miniaturization, integration with different light sources, and exploring higher-density metasurfaces to improve spatial resolution.

The current system's spatial resolution is limited by the number of point cloud projections. Although the system shows robustness to varying reflectivity, significant variations in surface properties might still challenge the accuracy. Further research is needed to explore the system's performance under diverse environmental conditions.