CVPR 2011 Structured light · Detailed example
Concave Bowl on a Marble Slab
The challenge
One scene.
Two illumination effects.
The concave bowl and translucent marble slab create different problems for structured-light decoding. Inter-reflections affect low-frequency patterns, while subsurface scattering blurs high-frequency patterns. Complementary codes are needed to handle both effects.
01 / Projected patterns
Why a single pattern family can fail
02 / Reconstruction comparison
Combining codes reduces reconstruction errors
Conventional Gray codes suffer from inter-reflections. Modulated phase shifting explicitly separates direct and global illumination, but the low direct component on the marble slab leads to poor signal-to-noise ratio. The code ensemble produces significantly fewer errors in this example.
Patterns for
inter-reflections
Inter-reflections cause errors in patterns with low spatial frequencies. To avoid these errors, patterns containing only high spatial frequencies are needed. Conventional Gray codes and phase-shifting patterns span a range of frequencies; simple logical operations can instead construct codes containing only high frequencies. The example below uses a V-groove scene.
Patterns for
scattering and defocus
Local effects, including subsurface scattering and defocus, blur the incident illumination. These effects require low-frequency patterns. Tools from combinatorial mathematics are used to design binary patterns with high minimum stripe width. The figure compares the resulting stripe-width distributions.
Handling
multiple effects
Most scenes do not reveal their global illumination behavior in advance and may contain both long-range inter-reflections and local effects. Projecting an ensemble of codes allows a consistency check across the resulting depth estimates.
For example, four codes produce four depths for each scene point: two sets of logical codes optimized for inter-reflections and two sets optimized for local effects. Incorrect codes tend to produce differing errors. When two estimates agree, the method treats that agreement as strong evidence for the correct depth.
