Domecam: Multi-Context Camera Systems

• Domecam is a polysemous term defining camera systems that leverage dome geometries to extract latent physical properties in fields ranging from neutrino detection to tactile sensing.
• These systems rely on meticulous geometric and optical calibration methods to convert digital images into measurements of ice properties, optical turbulence, contact forces, or refractive distortions.
• Despite diverse applications, Domecam implementations share challenges such as model approximations, interface limitations, and the need for forward/inverse imaging transformations.

Domecam is a context-dependent term used in several research literatures for camera-based systems associated with a dome geometry, a dome enclosure, or a dome optical interface. In the cited work, it denotes at least four distinct technical classes: a camera-based calibration subsystem inside IceCube digital optical modules, a dome-mounted all-sky-adjacent meteor camera on the Subaru Telescope enclosure, a pupil-plane scintillation instrument for measuring dome turbulence in a telescope, and a miniaturised dome-shaped tactile sensor; related literature on spherical underwater dome ports provides a further geometric meaning for dome-based camera systems (Tönnis et al., 2023, Tanaka et al., 1 Aug 2025, Kornilov et al., 2 Sep 2025, Althoefer et al., 2023, She et al., 2021). This multiplicity is important because the shared name does not imply a shared measurement principle: depending on context, a Domecam may be a forward radiative-transfer imager, a livestream meteor monitor, a scintillation-based optical-turbulence monitor, a vision-based tactile fingertip, or a refractive camera system behind a spherical port.

1. Terminological scope and common design motifs

Across these usages, Domecam denotes an instrument in which camera observations are coupled to a structured physical interface: the refrozen “hole ice” and surrounding glacial ice in IceCube, the telescope dome and night sky in meteor monitoring, the telescope pupil and near-field turbulence in dome-seeing studies, the hemispherical elastomer in tactile sensing, or the spherical glass dome in underwater imaging (Tönnis et al., 2023, Tanaka et al., 1 Aug 2025, Kornilov et al., 2 Sep 2025, Althoefer et al., 2023, She et al., 2021). The commonality is therefore architectural rather than disciplinary: each system uses imaging to infer latent physical properties that are not directly accessible by conventional point sensing.

The term is not standardized across fields. In neutrino-detector instrumentation, “DOM camera” or “Domecam” refers to the IceCube Upgrade camera system integrated into digital optical modules (Tönnis et al., 2023). In observational astronomy, “domecam” may refer either to a dome-mounted outreach-and-science meteor camera on a telescope enclosure (Tanaka et al., 1 Aug 2025) or to an instrument measuring optical turbulence inside the dome by pupil-conjugate scintillation (Kornilov et al., 2 Sep 2025). In robotics, the term denotes a dome-shaped camera-based tactile sensor (Althoefer et al., 2023). A plausible implication is that any encyclopedic treatment must be polysemous rather than device-specific.

2. IceCube Upgrade Domecam: in-situ optical calibration in glacial ice

In the IceCube Upgrade, the camera-based calibration system is integrated into new digital optical modules deployed on seven densely instrumented strings, with DOMs spaced vertically by 3 m between depths of 2160 m and 2430 m below the surface (Tönnis et al., 2023). The camera system, developed together with an LED illumination system at Sungkyunkwan University, is installed in almost every new DOM to perform in-situ optical calibration. Its targets are the refrozen drill-hole “hole ice” and the surrounding bulk glacial ice, both of which contribute materially to detector systematics.

The hardware description is explicit. The actual sensor has a native resolution of 1312 × 979 pixels with a Bayer color mosaic arranged in 2 × 2 RGGB groups. Because the LEDs are predominantly blue, simulations typically use 500 × 500 pixels to match the order of the number of blue pixels, and the camera field-of-view is restricted to 90° for the pinhole model (Tönnis et al., 2023). The framework can simulate generic light sources with Gaussian angular emission profiles, with example simulations using a profile width of 22.5°, and LED orientation relative to the camera is treated as an explicit variable; example orientations include 240°, 270°, 290°, and 320°.

The principal calibration goals are spatially resolved measurements of hole ice morphology and optical properties, bulk-ice absorption and scattering between strings, and anisotropy relevant to IceCube’s optical ice model such as SPICE (Tönnis et al., 2023). The paper emphasizes bubble-rich columns and refrozen layers in hole ice, including scattering and absorption lengths within the column, because such structures remain a major source of systematic uncertainty in several IceCube analyses. The dense Upgrade geometry also permits adjacent-string imaging of LED light cones over “few 10 m” baselines.

Quantitative interpretation is mediated by CamSim, a Python wrapper around the PPC photon propagation code with modifications for camera and LED simulation (Tönnis et al., 2023). PPC, written in C/CUDA, handles photon transport through layered or homogeneous media; CamSim configures geometry, sources, and cameras, converts PPC output into synthetic images, and adds camera noise. The simulation can include a vertical “column” with distinct optical properties representing refrozen hole ice, passing parameters such as radius, optical lengths, and offsets to PPC. A cited example uses a 3.3 cm-diameter bubble column with a 5 cm scattering length and a 10 cm offset from the camera optical axis, viewed downward with a 90° field of view.

The forward model combines emission, transport, and imaging. The attenuation law is written as

$$I(d,\lambda)=I_0(\lambda)\exp[-\kappa(\lambda)d], \quad \kappa(\lambda)=a(\lambda)+b(\lambda),$$

with scattering optionally represented by a Henyey–Greenstein phase function

$$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$

The camera projection is expressed in pinhole form,

$$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$

and the representative pixel-intensity model includes direct, scattered, and surface-reflection terms before conversion to counts (Tönnis et al., 2023). The framework also notes Snell’s law and Fresnel reflectance at interfaces such as ice–air bubbles and ice–DOM boundaries.

The empirically measured camera-noise model is a critical component. Laboratory characterization on 20 cameras at −40 °C with 3.7 s exposures gave a darkness baseline of approximately 240 counts, and the pixel-to-pixel variation follows approximately

$s \approx \sqrt{n-240},$

where $n$ is the pixel count under illumination (Tönnis et al., 2023). This model is applied to simulated images and then used in pixel-level likelihoods for parameter extraction. The cited analysis direction is likelihood-based fitting of cone shape, brightness falloff, near-field wall reflections, and hole-column signatures to infer scattering length, absorption length, hole-ice radius and offset, and LED orientation.

Current limitations are explicitly stated. Bulk layering is neglected for camera runs because the camera-visible range is limited to a few tens of meters; lens effects beyond the pinhole model are still being added; CLsim integration is planned; exact LED wavelengths and comprehensive spectral calibration are not provided (Tönnis et al., 2023). These constraints place the present Domecam primarily in the category of local, forward-modeled calibration rather than fully spectrally resolved inversion.

3. Domecam as a dome-turbulence instrument in observational astronomy

A second meaning of Domecam appears in telescope-site instrumentation, where it denotes a pupil-plane scintillation camera designed to quantify optical turbulence inside the telescope dome and convert it into delivered image quality for the 2.5-m Caucasian Mountain Observatory telescope (Kornilov et al., 2 Sep 2025). Here the physical observable is not a scene image but intensity fluctuation of a bright star in a plane conjugated to approximately −2 km below the telescope entrance pupil.

The physical principle is generalized SCIDAR-like near-field layer isolation. Under weak scintillation, phase perturbations $\phi(x)$ in a layer at distance $z$ evolve into pupil-plane amplitude fluctuations with Fourier transform

$$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$

The characteristic Fresnel radius is

$r_F=\sqrt{\lambda z}.$

By conjugating the detection plane to −2 km, Domecam increases the Fresnel phase for near-field layers; for $\lambda \approx 650\,\mathrm{nm}$ this gives $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$0, matched to the instrument’s spatial sampling (Kornilov et al., 2 Sep 2025). High-altitude layers are discriminated by aperture filtering and by temporal separation of autocovariance peaks under frozen-flow translation.

The instrument shares the F/8 Nasmyth N2 beam with the Speckle Polarimeter via a dichroic that reflects $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$1 to Domecam and transmits $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$2 to SPP (Kornilov et al., 2 Sep 2025). A KS-19 filter, a 4 mm field diaphragm corresponding to approximately 40 arcsec on sky, and a Fabry lens convert the focal plane to a 3.38 mm collimated beam. The plane 3.65 mm behind the exit pupil is optically conjugated to −2 km below the entrance pupil. The detector is a Prosilica GC650 CCD with 659 × 493 pixels of 7.4 $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$3m, operated with 2×2 binning; each binned pixel maps to 1.1 cm at the entrance pupil. Typical operation uses 2–4 ms exposures, 100 Hz frame rate, and 60 s series on bright stars with $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$4 mag.

The data-processing chain is highly structured. A mean bias frame is subtracted; a pupil mask is estimated on the 60 s average using Otsu thresholding; frames are normalized by total flux to remove extinction; relative fluctuations $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$5 are formed from 1 s estimates of the long-exposure mean pupil; and delayed 2D autocovariances $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$6 at $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$7 are computed using Wiener–Khinchin and FFTW3 (Kornilov et al., 2 Sep 2025). Dome optical turbulence appears as a narrow central peak at the −2 km conjugate plane, while windy high-altitude layers produce off-center peaks that move with delay.

The main scalar output is the dome optical-turbulence power $$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$8,

$$p(\theta)=\frac{1}{4\pi}\frac{1-g^2}{(1+g^2-2g\cos\theta)^{3/2}}.$$9

where the digital-filter impulse $$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$0 approximates $$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$1 over the filter support (Kornilov et al., 2 Sep 2025). This synthetic scintillation index estimates a quantity proportional to $$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$2 for near-field layers. For each series, four delayed $$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$3 values are fit by an exponential to extrapolate to zero-delay $$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$4 and a decorrelation timescale $$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$5.

Conversion to delivered image quality is performed through a von Kármán phase model. The phase power spectral density is

$$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$6

and the long-exposure optical transfer function is

$$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$7

Using simultaneous SPP measurements and MASS–DIMM free-atmosphere integrals, the authors fit the dome outer scale by Bayesian inference, fixing $$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$8 and obtaining $$\mathbf{x}_{\text{image}} = K[\,R \mid t\,]\mathbf{X}_{\text{world}},$$9 with an additional DIQ scatter $s \approx \sqrt{n-240},$0 (Kornilov et al., 2 Sep 2025). With this constraint, each measured $s \approx \sqrt{n-240},$1 is converted into a dome phase structure function and then into a DIQ increment.

The headline result is operationally significant: at 500 nm and zenith, dome turbulence increases the median expected delivered image quality from $s \approx \sqrt{n-240},$2 to $s \approx \sqrt{n-240},$3 (Kornilov et al., 2 Sep 2025). Dome optical-turbulence power varies from approximately $s \approx \sqrt{n-240},$4–$s \approx \sqrt{n-240},$5 under near-equilibrium thermal conditions to approximately $s \approx \sqrt{n-240},$6 when the primary mirror is about $s \approx \sqrt{n-240},$7 warmer than the dome air. The dominant driver is temperature disequilibrium between M1 and dome air; wind affects the decorrelation time more clearly than the turbulence amplitude itself.

This sense of Domecam should be distinguished from other dome-seeing instruments. The DIMSUM system at the Rubin Auxiliary Telescope addresses the same general problem—local refractive-index fluctuations inside an enclosure—but uses strobed multisource differential image motion rather than pupil-conjugate scintillation (Kurmus et al., 2023). DIMSUM places a modified Canon 5D Mark IV camera about 2 feet above the floor and 6.5 m from a 16-source fiber-fed flash panel, then uses 125 $s \approx \sqrt{n-240},$8s xenon strobe exposures and pairwise differential centroiding across $s \approx \sqrt{n-240},$9 source pairs to construct dome-seeing proxies (Kurmus et al., 2023). The relation

$n$0

is given as context, but the authors avoid a global $n$1 inversion and instead use empirical DIM-versus-separation metrics. The comparison clarifies that “Domecam” in astronomy may denote either a specific scintillation camera (Kornilov et al., 2 Sep 2025) or, more broadly, a dome-local optical sensor class exemplified by DIMSUM (Kurmus et al., 2023).

4. Domecam as a dome-mounted meteor and outreach camera

A third usage appears in wide-field meteor observation at the Subaru Telescope, where the Subaru-Asahi StarCam is described as a dome-mounted high-sensitivity live-streaming camera system and summarized as a “domecam” (Tanaka et al., 1 Aug 2025). It is mounted on the handrail of the catwalk on the fixed portion of the 44 m-diameter Subaru Telescope enclosure at Maunakea, at WGS84 latitude 19.8256°, longitude −155.4758°, with camera altitude 4153 m and physical mounting approximately 13 m above ground.

The field of view is 70° × 40°, oriented approximately due east, with FoV center at azimuth $n$2 and elevation $n$3 (Tanaka et al., 1 Aug 2025). The horizon lies about 5% above the bottom edge, and summit structures are intentionally included; the obscuration rate is approximately 23%, leaving a sky fraction of 77%. Visible facilities include Keck I and II, IRTF, CFHT, Gemini North, and UH88. The site context is central to performance: more than 50% photometric nights, more than 70% usable nights, and Subaru records showing 18.3% dome-closure nights per year over 2021–2023, consistent with approximately 70% usability for meteor work.

The hardware stack is specified in detail. The camera body is a Sony FX3 with a 10.2 megapixel full-frame backside-illuminated CMOS sensor, native ISO 80–102400, and expanded sensitivity to ISO 409600 (Tanaka et al., 1 Aug 2025). The lens is a Sony FE 24 mm F1.4 GM. A soft filter, Kenko PRO1D Clear Filter, is used to mitigate saturation of bright objects in the 8-bit livestream, improve constellation visibility, and aid relative magnitude estimation. Current livestream settings include Program Auto exposure, ISO AUTO [80–409600], exposure compensation +0.7 EV, Auto Slow Shutter ON, manual focus, Creative Look “VV”, white balance at 4500 K, and 2160p HDMI output.

Real-time performance is characterized by an effective frame rate of 15–30 fps and a single-frame limiting magnitude near $n$4 under dark clear Maunakea conditions (Tanaka et al., 1 Aug 2025). A representative operating point is ISO 409600 with shutter speed $n$5 at 30 fps, where most stars brighter than $n$6 and many fainter than 8 mag are detected. Streaming is continuous via YouTube since April 3, 2021, typically with more than 100 concurrent nighttime viewers, while the 4K stream is transported over more than 30 m using optical HDMI rather than Wi-Fi, which is prohibited at the summit.

Photometric and astrometric workflows convert the outreach stream into quantitative data products. The photometric pipeline splits 1 s video into frames, median-combines them, solves astrometry with astrometry.net, detects stars with DAOStarFinder, cross-matches to the Hipparcos Main Catalog, and applies vignetting correction (Tanaka et al., 1 Aug 2025). The paper notes 30–40% flux loss near the FoV edge, corresponding to approximately 0.4–0.5 mag, based on dense-fog flat-field proxy data. For meteor photometry, a local linear fit between instrumental counts and catalog magnitudes is derived from frames immediately before and after the event, subject to the limitation that bright-star saturation restricts the usable calibration range. The standard magnitude–flux relation

$n$7

is given as applicable.

Astrometric calibration produces RA/Dec (J2000) coordinates for meteor start and end points (Tanaka et al., 1 Aug 2025). Assuming a 3840 × 2160 stream, the pixel scales are

$n$8

$n$9

A practical output layer is provided by the “Hawaii Maunakea Meteor Database,” which automatically detects meteors from the livestream and publishes start/end positions, shower classification, approximate event time, average velocity, crude brightness estimate, and color; the database contained approximately 583,000 entries as of April 2025 (Tanaka et al., 1 Aug 2025).

The documented scientific outcomes include detection of the new Arid meteor shower in 2021, identification of a sub-peak activity in the Gamma-Perseid meteor shower in 2021, detection of the 2022 Tau-Herculid outburst, confirmation of Andromedid activity in 2021, and multiple detections of meteor cluster phenomena (Tanaka et al., 1 Aug 2025). Planned extension to a Maunakea–Mauna Loa two-station system with approximately 34 km baseline is intended to enable triangulation and orbital determination. In this usage, domecam is a hybrid outreach-and-science platform whose scientific utility derives directly from continuous high-sensitivity video under unusually favorable site conditions.

5. Domecam in tactile sensing and robotic manipulation

In robotics, Domecam denotes a miniaturised dome-shaped camera-based multi-modal tactile sensor designed to overcome limitations of flat GelSight-like sensors, especially bulkiness, difficulty measuring net forces because of elastomer hysteresis and nonlinearity, and reduced performance for point-like edge and corner contacts (Althoefer et al., 2023). The device adopts a curved silicone hemisphere and a compact spring-loaded structure so that a single internal camera can observe both local deformation and rigid-body deflection.

The complete sensor envelope is 24 mm × 24 mm × 26 mm, with a 20 mm-diameter dome elastomer (Althoefer et al., 2023). The housing uses transparent VeroClear upper and lower platforms. Force sensing is implemented with compression springs of length 5 mm, outer diameter 3 mm, and wire diameter 0.3 mm, retained by twelve 3 mm-diameter magnets. An acrylic sheet above the lower platform carries four white fiducials for tracking rigid-body deflection of the spring stack. The maximum measured load force is 17 N.

The optical stack consists of a 7.5 mm-diameter ultra-mini CMOS color UVC camera with a 160° wide-angle lens mounted at the base, together with a 0.8 mm-thick PCB carrying LUXEON CZ RGB LEDs tilted by 85° relative to the elastomer’s bottom plane (Althoefer et al., 2023). Rough-surface colored tapes diffuse and color-separate the beams, improving illumination uniformity on the curved Lambertian surface while avoiding the long light paths typical of flat GelSight designs. Black fiducial dots on the dome are produced by pad printing because oil-based ink is difficult to print directly on curved silicone.

The sensing principle is explicitly bimodal. Net force and torque are encoded by spring compression:

$\phi(x)$0

and, in matrix form, the generalized wrench is $\phi(x)$1 (Althoefer et al., 2023). In practice, Domecam does not observe individual spring compressions directly; instead, the four white fiducials on the acrylic plane are tracked and the plane’s 6D pose $\phi(x)$2 is recovered by SolvePnP from their known 3D coordinates. Because the mechanism is approximately linear over the operating range, the multi-axis wrench is then calibrated as a linear function of pose, written as $\phi(x)$3.

Local deformation and force distribution are derived from marker motion on the dome. After undistortion, sparse optical flow between frames yields a displacement field on the spherical surface. The dome is parameterized as

$\phi(x)$4

with camera pixels mapped to $\phi(x)$5 through a pinhole model (Althoefer et al., 2023). The authors use displacement magnitudes and directions directly as proxies for local force distribution rather than performing a full constitutive inversion. The shaded contact imprint under RGB illumination provides contact-geometry information analogous to retrographic sensing.

The processing pipeline includes camera calibration for the intrinsic matrix $\phi(x)$6 and distortion coefficients $\phi(x)$7, Gaussian denoising, masking, thresholding for black and white fiducials, morphological cleanup, sparse optical flow, and SolvePnP (Althoefer et al., 2023). The system runs at 30 Hz in Python/OpenCV. For learning-based inference, the paper uses a ResNet-18 backbone followed by a bidirectional GRU. Ten evenly sampled frames per video are processed, the fully connected layer outputs a 128-dimensional feature per frame, and the last five latent features feed a sigmoid classifier. Training uses SGD with learning rate 0.001 for 50 epochs on 500 videos of 60 frames each, giving 24,000 training frames and 6,000 validation frames. Reported hardness-classification performance is 99.054% precision with validation cross-entropy approximately 0.031 (Althoefer et al., 2023).

This Domecam differs conceptually from the optical and astronomical systems. Its “dome” is the tactile interface itself rather than an enclosure or detector housing. Nevertheless, the same structural theme persists: a compact imaging system observes a dome-mediated transformation of a physical field, here converting local contact mechanics and global wrench into visual signals.

6. Refractive dome-port cameras and the geometric meaning of Domecam

A further technical meaning arises in underwater imaging through spherical dome ports. In this literature, the relevant object is not explicitly named “Domecam” in the source paper, but the synthesis identifies dome-port camera systems as a coherent dome-camera class (She et al., 2021). The motivation is that spherical glass domes tolerate high pressure, preserve wide field of view, and eliminate refraction if the camera entrance pupil is located exactly at the sphere center. When decentered, however, the system no longer follows the pinhole model.

The core geometric result is that a decentered camera behind a spherical dome becomes an axial camera, even for thick domes used in deep-sea exploration (She et al., 2021). Let the dome center be the world origin $\phi(x)$8, the camera decentering vector be $\phi(x)$9, and the in-air projection matrix be

$z$0

The refraction axis is the line through $z$1 and $z$2, and the image of that axis defines the refraction center $z$3. For a 3D point $z$4, the as-in-air projection $z$5, the actual refracted projection $z$6, and $z$7 are collinear in the image:

$z$8

This reduces forward projection to a one-dimensional search along the line connecting $z$9 and $$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$0.

Refraction is governed by Snell’s law,

$$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$1

or, in vector form,

$$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$2

where $$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$3 (She et al., 2021). For thin domes, the authors derive a sixth-degree polynomial for forward projection in the refraction plane; for thick domes, an iterative 1D ray-tracing search is recommended because closed-form projection becomes numerically fragile.

A distinctive contribution is the direct, non-iterative estimation of the center of refraction from underwater chessboard images without requiring knowledge of exact refractive indices, dome thickness, or scene depth (She et al., 2021). If $$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$4 denotes chessboard coordinates and $$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$5 the refracted image points, the as-in-air coordinates satisfy $$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$6, while each refracted point obeys

$$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$7

Defining $$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$8, one obtains

$$\tilde{\chi}(\mathbf{f})=\tilde{\phi}(\mathbf{f})\sin(\pi \lambda z f^2).$$9

from which $r_F=\sqrt{\lambda z}.$0 is estimated linearly and $r_F=\sqrt{\lambda z}.$1 is recovered as the left null vector satisfying

$r_F=\sqrt{\lambda z}.$2

The axis direction then follows as

$r_F=\sqrt{\lambda z}.$3

Experimental highlights in the source include synthetic validation, rendered deep-sea dome experiments, and real-world calibration of a dome with radius 50 mm and thickness 7 mm (She et al., 2021). In the real-world case, pure underwater calibration yielded $r_F=\sqrt{\lambda z}.$4 and average $r_F=\sqrt{\lambda z}.$5 with reprojection RMSE approximately 0.32 px. In an AUV camera after mechanical adjustment, residual decentering was approximately $r_F=\sqrt{\lambda z}.$6 with calibration residual approximately 0.299 px.

This literature broadens the conceptual range of Domecam. Here the dome is a refractive optical boundary, and the camera’s central task is geometric reconstruction under spherical refraction rather than local environment imaging. A plausible implication is that the shared term “dome” links distinct systems through boundary geometry and calibration problems rather than through a unified sensing modality.

7. Comparative interpretation, limitations, and recurring methodological themes

The most important misconception to avoid is that Domecam names a single instrument family. In the cited literature it labels substantially different systems with different observables, forward models, and calibration targets (Tönnis et al., 2023, Tanaka et al., 1 Aug 2025, Kornilov et al., 2 Sep 2025, Althoefer et al., 2023, She et al., 2021). IceCube Domecam estimates local optical properties in ice; the CMO Domecam estimates near-field turbulence power and DIQ increments; Subaru’s domecam records meteors and related sky phenomena; the tactile Domecam infers wrench, deformation, and contact geometry; underwater dome-port cameras solve a refractive projection problem.

Despite that heterogeneity, several recurring methodological themes are visible. First, all variants depend on explicit geometric calibration: DOM and LED poses in CamSim, field-of-view and astrometric solutions in StarCam, pupil conjugation and pixel-to-pupil mapping in telescope Domecam, camera undistortion and SolvePnP in tactile Domecam, and refraction-center estimation in underwater dome geometry (Tönnis et al., 2023, Tanaka et al., 1 Aug 2025, Kornilov et al., 2 Sep 2025, Althoefer et al., 2023, She et al., 2021). Second, all variants translate images into physically meaningful latent variables through forward or inverse models rather than direct visual inspection. Third, each variant is limited by a specific mismatch between the idealized model and the real interface: neglected bulk layering in IceCube, 8-bit dynamic-range limits and vignetting in Subaru StarCam, model dependence on the von Kármán spectrum and outer scale in dome-turbulence monitoring, spring nonlinearity and illumination drift in tactile sensing, and decentering-induced deviation from perspective projection in underwater domes.

The reported future directions are similarly domain-specific. IceCube plans CLsim inclusion, additional lens models, and continued in-situ validation with joint fits to flasher data (Tönnis et al., 2023). Telescope dome-turbulence work proposes multi-wavelength operation, enhanced digital filters, machine-learning classifiers, and extension to other telescopes (Kornilov et al., 2 Sep 2025). The Subaru system aims at remote power control, multi-station expansion, refined flat-fielding, and broader Hawaiian network development (Tanaka et al., 1 Aug 2025). Tactile Domecam points toward richer end-to-end multimodal fusion on raw images (Althoefer et al., 2023). Underwater dome-port geometry suggests richer targets and better-coupled intrinsic–refractive calibration as future work (She et al., 2021).

Taken together, Domecam is best understood as a polysemous technical label for camera systems in which dome-associated geometry or enclosure physics is central to measurement. What unifies the usages is not hardware homogeneity but the systematic conversion of camera data into calibrated estimates of otherwise hidden physical structure.