Research map / 3D vision / Depth and stereo
Dense Dispersed Structured Light for Hyperspectral 3D Imaging of Dynamic Scenes
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Dense Dispersed Structured Light (DDSL), an affordable active-stereo system (RGB stereo cameras + RGB projector with diffraction grating film) that reconstructs hyperspectral images and depth of dynamic scenes from 8 spectrally multiplexed patterns at 6.6 fps.
Problem
Hyperspectral 3D imaging methods either need minutes of acquisition with hundreds of projected patterns or rely on expensive bulky hyperspectral video systems, so they cannot handle dynamic scenes.
Approach
Spectrally multiplexed DDSL line patterns projected through a diffraction grating film reduce the pattern count to eight; a data-driven dispersion-aware backward mapping, black-pattern optical flow for motion compensation, and a per-pixel regularized inversion recover hyperspectral reflectance and depth from stereo RGB images.
Key results
- Achieves 15.5 nm spectral FWHM versus 18 nm for Shin et al. and 40 nm for Li et al.
- Average depth error of 4 mm (max 8 mm) on a 3D-printed stair object, with ground truth from binary-coded structured light.
- Uses 8 DDSL patterns for 6.6 fps dynamic capture, versus about 10 minutes acquisition for Shin et al. (about 4000x speedup claimed).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use DDSL instead of Li et al. [13] | Hyperspectral 3D imaging · 10 nm-FWHM spectral bandpass filters | spectral FWHM (nm): 40 nm -> 15.5 nm | Helped |
| Use DDSL instead of Shin et al. [19] | Hyperspectral 3D imaging · 10 nm-FWHM spectral bandpass filters | spectral FWHM (nm): 18 nm -> 15.5 nm | Helped |
| Use DDSL instead of Shin et al. [19] | Scene acquisition · Scene acquisition | acquisition time: 10 minutes -> 0.15 seconds | Helped |
| Set loffset to 60 instead of the chosen 40 (ablation) | DDSL · DDSL pattern parameter ablation | spectral FWHM [nm]: 15.5 -> 15.5 | No clear effect |
| Set loffset to 20 instead of the chosen 40 (ablation) | DDSL · DDSL pattern parameter ablation | spectral FWHM [nm]: 15.5 -> 17 | Hurt |
| Set lwidth to 20 instead of the chosen 5 (ablation) | DDSL · DDSL pattern parameter ablation | spectral FWHM [nm] | Hurt |
| Set lwidth to 1 instead of the chosen 5 (ablation) | DDSL · DDSL pattern parameter ablation | spectral FWHM [nm]: 15.5 -> 50 | Hurt |
| Set lshift to 10 instead of the chosen 5 (ablation) | DDSL · DDSL pattern parameter ablation | spectral FWHM [nm] | Hurt |
| Set lshift to 2 instead of the chosen 5 (ablation) | DDSL · DDSL pattern parameter ablation | spectral FWHM [nm]: 15.5 -> 15.5 | No clear effect |
Limitations
- Frame rate limited to 6.6 fps by slow software synchronization between camera and projector, restricting to low-speed motion.
- Limited diffraction efficiency restricts the usable range of scene positions.
- Spectral FWHM of 15.5 nm is broader than the 10 nm filters, and spectral accuracy degrades with larger line width or offset.
Experimental setup
Datasets: spectralon calibration captures ColorChecker chart 3D-printed stair object real and artificial oranges
Metrics: spectral FWHM (nm) depth MAE (mm) acquisition time frame rate (fps)
Baselines: Shin et al. dispersed structured light Li et al. projector-camera hyperspectral 3D Kim et al. laser scanner with CASSI PSF-engineering methods [18] [4] Heist et al. active stereo with hyperspectral video
Builds on: dispersed structured light (Shin et al.) RAFT-Stereo RAFT binary-coded structured light
Closest papers by meaning
- Dispersed Structured Light for Hyperspectral 3D Imaging (CVPR 2024 · similarity 0.95)
- TurboSL: Dense Accurate and Fast 3D by Neural Inverse Structured Light (CVPR 2024 · similarity 0.86)
- HyperGS: Hyperspectral 3D Gaussian Splatting (CVPR 2025 · similarity 0.85)
- FRN: Fractal-Based Recursive Spectral Reconstruction Network (NeurIPS 2025 · similarity 0.84)
- Active Hyperspectral Imaging Using an Event Camera (CVPR 2025 · similarity 0.84)
- 3D Convex Splatting: Radiance Field Rendering with 3D Smooth Convexes (CVPR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.