Science - USA (2021-12-10)

at video rates, like those shown in Fig. 1 and
Fig. 2, are reconstructed by assuming that
the scene was illuminated with an ideally fo-
cused spot. However, if the actual projected
patterns are measured during the calibration
phase and do not change, a more sophisticated
reconstruction algorithm can be used to in-
corporate this information. We refer to this
hybrid method as TM-guided computational
imaging: Signal-to-noise ratio is maximized

by using the TM to concentrate power into

a focused spot, and regularized matrix in-

version is then used to account for residual

background speckle. Figure 4 depicts a series

of views of a static 3D image of a mannequin

head reconstructed from data recorded over

2 s using TM-guided computational imag-

ing. The depth error has been reduced to an

extent that enables the contours of the face

to be revealed. See supplementary materials,

section 10, for more details about this recon-

struction method.

Finally, we consider our work in the context

of alternative 3D imaging approaches. There

is an emerging class of clinical microendo-

scopes capable of recovering high-resolution

depth images within tissue using optical co-

herence tomography (OCT) ( 23 , 24 ). These

endoscopes range down to ~500mm in dia-

meter and typically deliver images with an

SCIENCEscience.org 10 DECEMBER 2021¥VOL 374 ISSUE 6573 1397

B

C

D

E

A

Fig. 2. Snapshots from depth-resolved movies at progressively increasing
scene depths.Left-hand column shows a direct camera image of each scene.
(A) Consecutive frames of a pendulum of ~1 m in length swinging adjacent
to a mannequin head and a RubikÕscubeat37to69cm(movieS2).Asmall
amount of rolling shutter effect is visible, causing the bob to appear to point in the
direction of motion as a result of the raster scan reaching the tip slightly later
in the swing than the base. (BandC) Author Daan Stellinga waving his hands

and shaking his head at 66 to 93 cm (movie S3). (DandE) Author

Peter Mekhail giving a high-five and dancing at 1 to 2.5 m (movie S4). Here,

human subjects wore eye protection, although we estimate the maximum

average power illuminating an exposed pupil is well below eye-safety

limits in our prototype (supplementary materials, section 11). The relative

time is given at the top of each frame. The frame number is given at the

bottom right of each frame.

RESEARCH | REPORTS