Dual-Camera Systems Overview

• Dual-Camera System (DCS) is an imaging configuration where two complementary cameras capture distinct measurements of the same scene to enhance image quality through calibrated fusion.
• It employs asymmetric sensing modalities, such as wide/telephoto pairings, alternating exposures, and spectral splits, to address challenges in HDR, super-resolution, and hyperspectral imaging.
• Successful DCS implementations rely on precise calibration, synchronization, and confidence-aware fusion to overcome alignment issues, occlusions, and hardware limitations.

Searching arXiv for recent and foundational papers on dual-camera systems relevant to the provided topic. to=arxiv_search.search 天天中彩票一等奖json code? to=arxiv_search.search ／久久json {"query":"dual-camera system arXiv smartphone HDR super-resolution hyperspectral ISP NeRF", "max_results": 10, "sort_by": "relevance"} to=arxiv_search.search ＿国产json {"query":"dual-camera system arXiv smartphone HDR super-resolution hyperspectral ISP NeRF", "max_results": 10} Dual-camera system (DCS) denotes an imaging configuration in which two cameras, or two coordinated sensing branches realized as cameras, capture complementary measurements of the same scene and are coupled by a reconstruction algorithm. In the literature, the complementarity may be geometric, radiometric, temporal, spectral, or optical: wide and telephoto cameras for super-resolution, stereo Bayer measurements for ISP, alternating and fixed exposures for HDR video, RGB and CASSI branches for snapshot hyperspectral imaging, long- and short-exposure pairs for low-light restoration, focused and defocused views for demoiréing, and shallow-DoF and deep-DoF views for refocus and all-in-focus NeRF synthesis (Wang et al., 2021, Rabiah et al., 2022, Zhang et al., 9 Jul 2025, Cai et al., 2023, Shekarforoush et al., 2023, Luo et al., 23 Apr 2025). Rather than a single algorithmic template, DCS is a systems concept: two acquisition channels are designed so that one branch supplies information that the other cannot capture reliably, and reconstruction hinges on calibration, alignment, and confidence-aware fusion.

1. Architectural patterns and sensing complementarity

A recurring architectural pattern is asymmetric sensing. In smartphone dual-camera super-resolution, the main wide-angle camera is treated as the “LR” input with a large field-of-view, while the telephoto camera is the “Ref” input with higher native resolution in a narrower field-of-view; the two images are true raw ISP outputs with real translation, rotation, scale, and color mismatch (Wang et al., 2021). In stereo ISP, the primary and secondary sensors are a rectified stereo pair producing two raw Bayer measurements, $R_L$ and $R_R$, that differ mainly by disparity and noise realization rather than by focal length (Rabiah et al., 2022). In HDR video DCS, one camera continuously acquires medium-exposure reference frames $L_m(t)$, while the other alternates between low and high exposures $L_l, L_h(t)$, so that temporal consistency is anchored to a fixed-exposure stream (Zhang et al., 9 Jul 2025).

Another major pattern is optical branching through a beamsplitter. Both DMDC and the self-supervised dual-camera hyperspectral system use a 50:50 beamsplitter to split one scene beam into an RGB or color branch and a CASSI-like spectral-compression branch (Cai et al., 2023, Xie et al., 2021). In these systems, the RGB path supplies spatial structure, while the coded-aperture dispersive path supplies compressed spectral information. This differs from smartphone wide/telephoto or wide/ultra-wide DCS, where the two sensors observe the scene from slightly different viewpoints with different focal lengths and apertures rather than through a common optical axis.

A third pattern is optical complementarity of depth of field and exposure. DC-NeRF and $\text{DC}^2$ exploit a main camera with shallower DoF and an ultra-wide camera with deeper DoF; the main branch preserves higher-fidelity detail at the focal plane, whereas the ultra-wide branch provides sharper reference content in out-of-focus regions (Luo et al., 23 Apr 2025, Alzayer et al., 2023). Dual-camera joint deblurring-denoising and mobile face deblurring use a long-exposure or AE-driven main shot that is cleaner but blurrier, paired with a short-exposure auxiliary shot or burst that is sharper but noisier (Shekarforoush et al., 2023, Lai et al., 2022).

These configurations show that DCS is fundamentally a co-design problem. The second camera is not merely redundant; it is assigned a sensing regime that is deliberately different from the first. A plausible implication is that the central design variable in DCS is not camera count alone, but the extent to which the pair spans complementary failure modes.

2. Calibration, synchronization, and acquisition constraints

Because the two channels are not identical, DCS pipelines begin with calibration. HDR video DCS uses intrinsic and extrinsic calibration via checkerboard following Zhang 2002 to remove distortion and register views; the two cameras are triggered together but operate asynchronously, and timestamp grouping forms each HDR input group (Zhang et al., 9 Jul 2025). DC-NeRF estimates intrinsics and per-view extrinsics with COLMAP, then aligns the ultra-wide image to the main camera with a global homography from SIFT, NM-Net, and RANSAC followed by RAFT-based optical flow and histogram matching (Luo et al., 23 Apr 2025). In view-transition fusion for wide/telephoto imagery, FlowFormer produces the original backward flow $F_{bwd}^{T\to W}$, after which the flow itself is modified so that the output lies in a mixed view rather than in the wide view (Cao et al., 2023).

Temporal synchronization can dominate system behavior. In the particle-image-velocimetry DCS of Hashimoto et al., two Nikon D70S still cameras are triggered with a user-set delay, while flash-sync signals are recorded to recover the actual exposure timing. The true inter-exposure interval is

$$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$

and the shutter-lag jitter is reported as approximately $1.4$ ms, which caused only $\sim 10\%$ of image pairs to coincide properly with the laser pulses in the prototype (Hashimoto et al., 2012). By contrast, the HDR video DCS explicitly avoids frame-level hardware synchronization and instead uses nearest timestamps, reflecting a different operating point in the design space (Zhang et al., 9 Jul 2025).

Acquisition constraints may also be formalized as optimization variables. In “Optimal HDR and Depth from Dual Cameras,” the total capture time is minimized subject to full radiance-range coverage, an overlap constraint for disparity estimation, and a per-frame SNR lower bound. The overlap constraint is defined on the radiance intervals seen by both cameras, which makes explicit that dual cameras do not automatically shorten HDR capture unless enough exposure overlap is retained for stereo matching (Chari et al., 2020). This result is significant because it turns a common engineering trade-off—dynamic-range coverage versus stereo overlap—into an explicit non-convex planning problem.

Across these systems, calibration is not ancillary. It determines whether the second measurement functions as usable signal or as structured interference.

3. Registration and fusion formulations

DCS registration methods are typically more structured than generic image alignment because they are tied to the sensing asymmetry. In dual-camera super-resolution, coarse matching is performed on $3\times 3$ feature patches with cosine similarity,

$R_R$0

followed by patch selection $R_R$1 and confidence $R_R$2. A small Spatial Transformer then predicts a local affine matrix $R_R$3 for each matched patch, yielding aligned reference features that are fused with learned confidence weighting (Wang et al., 2021). StereoISP uses a more classical stereo formulation: a disparity module $R_R$4 produces $R_R$5, and the right raw Bayer measurement is backward-warped to $R_R$6 before a joint demosaic-denoise network $R_R$7 reconstructs RGB (Rabiah et al., 2022).

Fusion operators likewise reflect the task. DCSR combines feature-space and image-space fusion: $R_R$8 and adds an aligned high-frequency residual extracted from the reference image (Wang et al., 2021). AWnet for high-spatiotemporal-resolution video computes a warped high-resolution reference, refines it with a dynamic $R_R$9 filter, and blends it with the upscaled low-resolution high-frame-rate input by a learned mask,

$L_m(t)$0

which is explicitly an adaptive weighting function in the pixel domain (Cheng et al., 2019). EAFNet for HDR video uses medium-exposure-dominated attention: queries come from reference features, keys and values from non-reference features, and cross-scale guidance is used before a DWT-based multiscale reconstruction stage (Zhang et al., 9 Jul 2025).

In hyperspectral DCS, fusion is often expressed as a joint inverse problem. DMDC stacks RGB and CASSI measurements in a combined linear model,

$L_m(t)$1

and solves

$L_m(t)$2

with an unrolled reversible dual-stream network (Cai et al., 2023). The self-supervised physics-informed system instead optimizes an untrained CNN per scene so that its hyperspectral output simultaneously satisfies the color-camera forward model $L_m(t)$3 and the CASSI forward model $L_m(t)$4 (Xie et al., 2021). The later TVDS framework retains the same dual-camera CASSI idea but imposes a convex total-variation subgradient similarity term guided by the RGB reference, together with an ADMM solver and explicit convergence conditions (Zhao et al., 13 Sep 2025).

A common thread is that DCS fusion is usually confidence- or physics-aware. The systems do not merely concatenate two images; they estimate which branch should dominate at which location, scale, wavelength, or time.

4. Principal application classes

The term DCS spans several distinct application families rather than a single canonical use case.

Application class	Camera roles	Representative papers
Super-resolution and zoom	wide/main $L_m(t)$5 telephoto reference	(Wang et al., 2021, Cao et al., 2023)
ISP, HDR, and HDR video	stereo Bayer pair; dual exposure streams	(Rabiah et al., 2022, Chari et al., 2020, Zhang et al., 9 Jul 2025)
Snapshot hyperspectral imaging	RGB/color branch $L_m(t)$6 CASSI branch	(Cai et al., 2023, Xie et al., 2021, Zhao et al., 13 Sep 2025)
Deblurring, denoising, demoiréing	long $L_m(t)$7 short exposure; focused $L_m(t)$8 defocused	(Shekarforoush et al., 2023, Lai et al., 2022, Dong et al., 5 Aug 2025)
Refocus, all-in-focus NeRF, and video acquisition	shallow-DoF $L_m(t)$9 deep-DoF; HSR-LFR $L_l, L_h(t)$0 LSR-HFR	(Alzayer et al., 2023, Luo et al., 23 Apr 2025, Cheng et al., 2019)
Experimental imaging	interleaved still cameras for PIV	(Hashimoto et al., 2012)

Within super-resolution and zoom, one line of work keeps the output in the wide-camera frame and attempts to align telephoto content into it, while another line reduces the ill-posedness of occluded regions by shifting the output itself toward a mixed view between the wide and telephoto cameras (Wang et al., 2021, Cao et al., 2023). This suggests that “where the output viewpoint should live” is itself a DCS design variable.

Within HDR and ISP, DCS serves two somewhat different purposes. StereoISP uses two raws to improve demosaicing and denoising via disparity-aware raw fusion (Rabiah et al., 2022), whereas HDR-oriented systems use dual cameras either to plan optimal exposure/ISO sequences for joint HDR and disparity recovery (Chari et al., 2020) or to maintain temporal consistency by dedicating one camera to a stable reference stream (Zhang et al., 9 Jul 2025).

The hyperspectral branch of the literature is especially heterogeneous. DMDC uses the RGB image to predict a scene-adaptive SLM mask for the CASSI arm (Cai et al., 2023). The physics-informed framework dispenses with supervised training and reconstructs by enforcing the optical forward models per scene (Xie et al., 2021). TVDS, in turn, emphasizes convex analysis and interpretable regularization, using RGB or panchromatic references to guide a total-variation subgradient prior (Zhao et al., 13 Sep 2025).

Restoration-oriented DCS methods also differ materially. In dual-camera joint deblurring-denoising, motion is sampled by a synchronized short-exposure burst and then injected into a non-blind deblurring network for the long exposure (Shekarforoush et al., 2023). In focused-defocused demoiréing, the defocused stream suppresses moiré while retaining coarse layout, so it becomes a guidance image rather than a texture source (Dong et al., 5 Aug 2025). In $L_l, L_h(t)$1 and DC-NeRF, the second camera is primarily a sharp reference for out-of-focus regions (Alzayer et al., 2023, Luo et al., 23 Apr 2025).

5. Empirical performance across tasks

Reported gains are substantial but task-specific. On CameraFusion for $L_l, L_h(t)$2 dual-camera super-resolution, “Ours ($L_l, L_h(t)$3)” reaches $L_l, L_h(t)$4 PSNR and $L_l, L_h(t)$5 SSIM, compared with $L_l, L_h(t)$6 for TTSR ($L_l, L_h(t)$7), $L_l, L_h(t)$8 for CSNLN, $L_l, L_h(t)$9 for RCAN, and $\text{DC}^2$0 for bicubic; on CUFED5 $\text{DC}^2$1, “Ours-$\text{DC}^2$2” reaches $\text{DC}^2$3 PSNR and $\text{DC}^2$4 SSIM versus $\text{DC}^2$5 for TTSR-$\text{DC}^2$6 (Wang et al., 2021). In wide/telephoto view-transition fusion, plain $\text{DC}^2$7 warping produces occlusion areas of $\text{DC}^2$8 on OPPO72 and $\text{DC}^2$9 on CameraFusion, whereas view transition reduces them to $F_{bwd}^{T\to W}$0 and $F_{bwd}^{T\to W}$1, respectively (Cao et al., 2023).

For raw-domain ISP, StereoISP reports a PSNR increase from $F_{bwd}^{T\to W}$2 dB to $F_{bwd}^{T\to W}$3 dB on KITTI 2015 and from $F_{bwd}^{T\to W}$4 dB to $F_{bwd}^{T\to W}$5 dB on drivingStereo when ground-truth sparse disparity maps are used (Rabiah et al., 2022). In the joint HDR+depth exposure-planning framework, the optimal sequence on six scenes attains approximately HDR-VDP-2 $F_{bwd}^{T\to W}$6 at $F_{bwd}^{T\to W}$7 s for HDR-only, compared with approximately $F_{bwd}^{T\to W}$8 at $F_{bwd}^{T\to W}$9 s for 3-shot single-camera bracketing; for joint HDR+depth, the optimal sequence is reported at $$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$0 s with VDP2 $$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$1 and disparity error $$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$2, compared with interleaved $$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$3 at $$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$4 s, VDP2 $$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$5, and disparity error $$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$6 (Chari et al., 2020). For stable HDR video, EAFNet reaches PSNR-$$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$7 $$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$8, SSIM-$$\Delta t_{\rm actual}=\Delta t_{\rm trig}+(\tau_B-\tau_A),$$9 $1.4$0, and HDR-VDP-2 $1.4$1 on the Kalantari dataset, and on self-captured HDR videos it reports $1.4$2-SSIM $1.4$3, $1.4$4, and runtime $1.4$5 ms per $1.4$6 patch (Zhang et al., 9 Jul 2025).

Hyperspectral DCS results are equally strong within their own benchmarks. On KAIST, DMDC-9stg reaches $1.4$7 dB and $1.4$8 SSIM, compared with best SOTA $1.4$9 dB and $\sim 10\%$0 for SST-LPlus; on ARAD_1K, DMDC-9stg reports MRAE $\sim 10\%$1 and PSNR $\sim 10\%$2 dB, compared with MRAE $\sim 10\%$3 and PSNR $\sim 10\%$4 dB for DAUHST-9stg (Cai et al., 2023). The untrained physics-informed dual-camera system reports average results around PSNR $\sim 10\%$5 dB, SSIM $\sim 10\%$6, and SAM $\sim 10\%$7 against iterative baselines on CAVE/ICVL/Harvard, and around PSNR $\sim 10\%$8 dB and SSIM $\sim 10\%$9 against supervised CNNs under standard splits (Xie et al., 2021).

In restoration tasks, the synchronized long/short-exposure deblurring-denoising system reports PSNR $3\times 3$0 dB, SSIM $3\times 3$1, and LPIPS $3\times 3$2 with $3\times 3$3 M parameters on synthetic dual-camera GoPro data (Shekarforoush et al., 2023). The mobile face-deblurring DCS adds $3\times 3$4 ms overhead per shot on Google Pixel 6 and reports NIMA $3\times 3$5 on a representative set, compared with $3\times 3$6 for the next best baseline (Lai et al., 2022). For focused-defocused video demoiréing, the dual-camera method reports $3\times 3$7 dB PSNR, $3\times 3$8 SSIM, and $3\times 3$9 LPIPS on DualSynthetic, and on DualSyntheticVideo it reports $R_R$00 dB, $R_R$01, $R_R$02, $R_R$03-MSE $R_R$04, and $R_R$05-SSIM $R_R$06 (Dong et al., 5 Aug 2025).

For depth-of-field–related tasks, $R_R$07 reports $R_R$08 for defocus deblurring, $R_R$09 for bokeh rendering, and $R_R$10 for image refocus, outperforming the baselines listed in its tables (Alzayer et al., 2023). DC-NeRF reports the best average among the listed methods over seven scenes, with $R_R$11 dB PSNR, $R_R$12 SSIM, and $R_R$13 LPIPS (Luo et al., 23 Apr 2025). In high-spatiotemporal-resolution video acquisition, AWnet reports $R_R$14 dB and $R_R$15 SSIM on Vimeo90K at $R_R$16, versus $R_R$17 for CrossNet (Cheng et al., 2019). In PIV, the dual-camera prototype achieves a $R_R$18 ms inter-pulse interval, spatial resolution $R_R$19, scale $R_R$20 mm/px, and velocity uncertainty $R_R$21 m/s (Hashimoto et al., 2012).

These results do not define a single DCS performance frontier. They show instead that dual-camera gains are largest when the complementary measurement is tightly coupled to the inverse problem: disparity for raw ISP, exposure asymmetry for HDR video, dynamic coding for CASSI, or DoF asymmetry for refocus and NeRF.

6. Limitations, misconceptions, and research directions

A frequent misconception is that adding a second camera automatically improves quality. The literature repeatedly shows that the improvement depends on alignment fidelity and overlap. In StereoISP, feeding the unaligned stereo pair into the network drops PSNR to $R_R$22 dB on KITTI, below the single-view baseline of $R_R$23 dB (Rabiah et al., 2022). In DCSR, the fidelity loss is reported to prevent fallback to SISR; without it, details vanish (Wang et al., 2021). In DMDC ablations on ARAD_1K, removing the RGB branch costs $R_R$24 dB, removing Cross-Attention costs $R_R$25 dB, and replacing the dynamic mask with fixed masks costs $R_R$26 dB (Cai et al., 2023). These results indicate that the second branch must remain both informative and actually used by the optimizer.

Parallax and occlusion remain structural difficulties. View-transition fusion is motivated precisely by the claim that keeping the output fixed in the wide view makes enhancement in occlusion areas ill-posed (Cao et al., 2023). HDR video DCS still reports residual parallax artifacts in extreme depth disparities, despite asynchronous capture robustness (Zhang et al., 9 Jul 2025). DC-NeRF notes that strongly defocused edges can mislead optical flow and that ultra-wide regions with extremely low texture may yield low-quality radiance (Luo et al., 23 Apr 2025). The mobile face-deblurring system requires an explicit occlusion mask because boundary artifacts appear where the ultrawide warp is invalid (Lai et al., 2022).

Hardware simplicity is also not guaranteed. The PIV prototype trades continuous-burst capability for higher resolution and sensitivity per exposure, but suffers from shutter-lag jitter and only two frames per measurement (Hashimoto et al., 2012). HDR video DCS requires two cameras and mechanical mounting (Zhang et al., 9 Jul 2025). Physics-informed hyperspectral reconstruction avoids labeled data but reports approximately $R_R$27 minutes for one $R_R$28 reconstruction (Xie et al., 2021), while DC-NeRF reports approximately $R_R$29 h per scene on a single RTX-3090 (Luo et al., 23 Apr 2025). Even mobile-oriented DCS pipelines incur nontrivial latency and memory overhead (Lai et al., 2022).

The trajectory of the field is therefore toward tighter hardware-software co-design rather than toward camera multiplicity alone. The papers explicitly suggest end-to-end joint training of stereo matching and ISP, model compression and quantization for mobile deployment, on-device FPGA or ASIC acceleration for HDR DCS, extension to three or more cameras, hyperspectral video, and additional depth-aware alignment strategies (Rabiah et al., 2022, Zhang et al., 9 Jul 2025, Cao et al., 2023, Xie et al., 2021). This suggests that future DCS research will be shaped less by the mere existence of a second sensor than by principled choices about viewpoint, exposure, coding, and optimization domain.