Alankar Kotwal, Anat Levin, and Ioannis Gkioulekas
We present a new imaging technique, swept-angle synthetic wavelength interferometry, for full-field micron-scale 3D sensing. As in conventional synthetic wavelength interferometry, our technique uses light consisting of two narrowly-separated optical wavelengths, resulting in per-pixel interferometric measurements whose phase encodes scene depth. Our technique additionally uses a new type of light source that, by emulating spatially-incoherent illumination, makes interferometric measurements insensitive to aberrations and (sub)surface scattering, effects that corrupt phase measurements. The resulting technique combines the robustness to such corruptions of scanning interferometric setups, with the speed of full-field interferometric setups. Overall, our technique can recover full-frame depth at a lateral and axial resolution of 5 microns, at frame rates of 5 Hz, even under strong ambient light. We build an experimental prototype, and use it to demonstrate these capabilities by scanning a variety of objects, including objects representative of applications in inspection and fabrication, and objects that contain challenging light scattering effects.
Micrometer-resolution depth sensing, our focus in this project, is important in biomedical imaging because biological features are often micron-scale, industrial fabrication and inspection of critical parts that must conform to their specifications, and robotics to handle fine objects. Active illumination depth sensing techniques such as lidar, structured light, and correlation time-of-flight (ToF) cannot provide micrometer axial resolution — so, we focus on interferometric techniques that can achieve such resolutions. The choice of interferometric illumination spectrum leads to techniques such as optical coherence tomography (OCT), which uses broadband illumination, and phase-shifting interferometry (PSI), which have complementary properties: the large unambiguous depth range of OCT, and the large axial resolution of PSI.
We consider synthetic wavelength interferometry (SWI), which operates between these two extremes: By using illumination consisting of two narrowly-separated optical wavelengths, SWI provides a controllable trade-off between the large unambiguous depth range of OCT, and the large axial resolution of PSI. SWI can achieve micrometer resolution at depth ranges in the order of hundreds of micrometers.
The choice of optical configuration results in full-field versus scanning implementations, which offer different trade-offs. Full-field implementations (such as the one in (a)) acquire 2D depth maps, offering fast operation and high lateral resolutions. However, they are very sensitive to effects such as indirect illumination (e.g., subsurface scattering), and aberrations in free-space optics that corrupt depth reconstruction. By contrast, scanning implementations (such as (b)) use beam steering to sequentially scan points in a scene. These offer robustness to depth corruption effects, but make acquiring megapixel depth maps with pixel-level lateral resolution impractically slow.
Gkioulekas et al. and Xiao et al. showed that using spatially incoherent illumination in a Michelson interferometer optically rejects depth corruption effects in full-field interferometry. They achieve spatial incoherence by replacing the point emitter in (a) with an area emitter such as an LED or a halogen lamp. Unfortunately, fundamental physics dictate that extending emission area of a light source is coupled with broadening its emission spectrum. This makes spatially-incoherent light sources incompatible with SWI, which requires narrow-linewidth dichromatic illumination.
To resolve this conundrum, in (c), we emulate a spatially-incoherent source suitable for SWI through time-division multiplexing: We use a galvo mirror to steer a narrow collimated beam of narrow-linewidth dichromatic illumination. We focus this beam through a relay lens at a point on the focal plane of the illumination lens. As we steer the beam direction, the focus point scans an area on this focal plane within exposure time of the camera. As this design sweeps the illumination angle within exposure, we term it swept-angle illumination, and its combination with full-field SWI swept-angle synthetic wavelength interferometry. This design is an adaptation of the Fourier-domain redistributive projector of Kotwal et al.
Here, we show scans of a variety of challenging scenes with microscopic depth features. We scan materials ranging from rough metallic (coins), to diffuse (music box), to highly-scattering (chocolate). Most of these scenes include fine features requiring high lateral and axial resolution (music box, business card, quarter). We compare scan results using swept-angle SWI (with bilateral or Gaussian filtering), versus conventional full-field SWI. In all scenes, the use of swept-angle illumination greatly improves reconstruction quality. The difference is more pronounced in scenes with strong subsurface scattering (music box, chocolate, soap). Even in metallic scenes where there is no subsurface scattering (coins), the use of swept-angle illumination still improves reconstruction quality, by helping mitigate aberration artifacts. Between bilateral versus Gaussian filtering, bilateral filtering helps preserve higher lateral detail.
Here, we show scans of a variety of challenging scenes with macroscopic depth features. As in the microscopic results, the use of swept-angle illumination greatly improves reconstruction quality. The difference is more pronounced in scenes with subsurface scattering (cup). Between bilateral versus Gaussian filtering, bilateral filtering helps preserve higher lateral detail.
We showcase results representative of applications of SWI in industrial fabrication and inspection: We use swept-angle SWI to scan a wing section from a decommissioned Boeing 747, to detect critical defects such as scratches and bumps from collisions, at axial and lateral scales of a couple dozen micrometers. We also use swept-angle SWI to scan a coin pattern 3D-printed by a commercial material printer on a translucent material.
To qualitatively assess the relative performance of swept-angle and scanning-based SWI we downsample the depth map from our technique. We then use joint bilateral upsampling to upsample the downsampled depth map back to its original resolution. We observe that, due to the sparse set of points scanning SWI can acquire within the exposure time, the reconstructions miss fine features such as the hair on the quarter and letters on the business card.
For an in-depth description of the technology behind this work, please refer to our CVPR 2023 paper. A pre-print and the supplementary material can be found here.
Alankar Kotwal, Anat Levin, and Ioannis Gkioulekas. "Swept-Angle Synthetic Wavelength Interferometry", CVPR 2023
We thank Sudershan Boovaraghavan, Yuvraj Agrawal, Arpit Agarwal, Wenzhen Yuan from CMU, and Veniamin V. Stryzheus, Brian T. Miller from The Boeing Company, who provided the samples for the experiments in Figure 1 of the paper. This work was supported by NSF awards 1730147, 2047341, 2008123 (NSF-BSF 2019758), and a Sloan Research Fellowship for Ioannis Gkioulekas.