Surround structured lighting: 3-D scanning with orthographic illumination

Douglas Lanman, Daniel E. Crispell, Gabriel TaubinCVIU 2009 · DOI

TL;DR

This paper introduces a novel 3D scanning system that uses a single orthographic structured light projector and a pair of mirrors to capture complete 360° object models in a single pass. This approach overcomes the limitations of traditional structured light systems, which require multiple scans or complex setups to achieve full object coverage, enabling faster and more efficient 3D reconstruction.

What Problem Does This Paper Address?

Traditional structured light 3D scanning systems, while cost-effective, typically rely on a single camera-projector pair. This setup can only capture the portion of an object that is both illuminated and visible to the camera. To achieve a complete 3D model, these systems usually require either rotating the object on a turntable or performing multiple scans from different viewpoints, followed by complex registration algorithms. This process is time-consuming and not suitable for capturing dynamic scenes. The paper addresses the challenge of obtaining a full 360° reconstruction efficiently and without the need for object manipulation or multiple scanning passes.

Methods

The core innovation lies in the system's architecture:

  • Orthographic Structured Light Projector: Instead of a standard projector, they use a combination of a commercial DLP projector and a Fresnel lens. This setup aims to produce parallel light rays, approximating an orthographic projection. This is crucial for the system's ability to illuminate from all sides.
  • Surround Illumination with Mirrors: A pair of planar mirrors are strategically placed around the object. These mirrors reflect the structured light patterns, effectively illuminating the object from multiple directions simultaneously.
  • Virtual Viewpoints: The mirrors create virtual viewpoints of the object. A single camera, positioned to view the object and its reflections in the mirrors, captures five distinct views: one direct view and four reflected views (two from first reflections and two from second reflections).
  • Gray Code Encoding: The system employs a standard Gray code sequence for structured lighting. This method assigns a unique code to each projected light plane, enabling unambiguous correspondence between camera pixels and projector rows. A key benefit highlighted is that even with illumination from multiple directions, each point on the object surface can still be assigned a unique Gray code sequence.
  • Calibration Procedures: The paper details comprehensive calibration procedures for the orthographic projector (aligning the DLP and Fresnel lens, and mapping projector pixels to 3D rays) and for the camera-mirror system (determining the precise position and pose of the mirrors relative to the camera). These procedures are critical for accurate reconstruction.

Datasets

Not discussed in the paper. The authors use custom setups and calibration patterns for their experiments.

Key Insights and Ideas

  • Orthographic Projection for Full Coverage: The realization that approximating orthographic projection, combined with mirrors, is key to illuminating an object from all sides with a single projector. This sidesteps the interference issues that arise when multiple standard projectors are used.
  • Mirror-Induced Virtual Views for Efficiency: Using mirrors to generate multiple viewpoints from a single camera is a clever way to increase the amount of information captured without adding more hardware or complex synchronization.
  • Unambiguous Gray Codes in Multi-Directional Illumination: The system's design ensures that even though a point might be illuminated from multiple directions, the Gray code sequence remains unambiguous, a critical factor for reliable 3D reconstruction.
  • Practical Orthographic Projector Construction: The paper provides concrete methods for building an orthographic projector using readily available components (DLP projector and Fresnel lens), making the technology more accessible.

Key Results

  • Complete 360° Reconstruction: The system successfully reconstructs nearly the entire surface of small objects, including the bottom, in a single scanning pass.
  • Reconstruction Accuracy: The prototype achieves a reconstruction error of less than 1 mm in practical situations.
  • Error Analysis: Reconstruction errors tend to increase with the number of mirror reflections, with standard deviations of 0.3743 mm for direct imaging, 0.4646 mm for one mirror reflection, and 0.6981 mm for two mirror reflections.
  • Mechanical Alignment is Crucial: The authors emphasize that mechanical alignment and the quality of optical components significantly impact accuracy, particularly for reflected views.

Limitations and Future Work

  • Limited Scanning Volume: The current reconstruction volume is constrained by the dimensions and focal length of the Fresnel lens.
  • Artifacts from Reflections and Optics: Visible distortion and decreased reconstruction accuracy are observed with reflected images, attributed to optical quality and mechanical alignment.
  • Occlusions: While not a major issue in their experiments, object-dependent occlusions can still impact reconstruction if an object blocks all illumination rays.
  • Future Work:
    • Larger Scanning Volume: Incorporating larger Fresnel lenses with longer focal lengths to increase the scanning volume.
  • Higher Quality Optics: Using better quality mirrors and lenses to reduce artifacts.
  • Real-time Capture: Implementing hardware synchronization to achieve real-time 3D reconstruction at the camera's maximum frame rate, suitable for dynamic scenes.
  • "Single-Shot" Patterns: Exploring single-image structured light patterns for even faster acquisition.
  • Improved Post-processing: Enhancing post-processing pipelines for better outlier removal and color blending, and to produce watertight models for rapid prototyping.
  • Sub-pixel Accuracy: Investigating structured light sequences that achieve sub-pixel accurate assignment of projector rows to camera pixels to reduce quantization noise.

Summary updated

BibTeX
@article{lanman2009surround,
  author = {Douglas Lanman and Daniel E. Crispell and Gabriel Taubin},
  title = {{Surround structured lighting: 3-D scanning with orthographic illumination}},
  journal = {Comput. Vis. Image Underst.},
  year = {2009},
  doi = {10.1016/j.cviu.2009.03.016},
}
Figure