Structured light 3D scanning in the presence of global illumination
Summary (2 min read)
1. Introduction
- Structured light triangulation has become the method of choice for shape measurement in several applications including industrial automation, graphics, human-computer interaction and surgery.
- Most structured light techniques make an important assumption: scene points receive illumination only directly from the light source.
- Imagine a robot trying to navigate an underground cave or an indoor scenario, a surgical instrument inside human body, a robotic arm sorting a heap of metallic machine parts, or a movie director wanting to image the face of an actor.
- The authors show that the types and magnitude of errors depend on the region of influence of global illumination at any scene point.
- Ery pixel, without any prior knowledge about the types of effects in the scene (Figure 1d).
3. Errors due to Global Illumination
- The type and magnitude of errors due to global illumination depends on the spatial frequencies of the patterns and the global illumination effect.
- For a scene point Si, its irradiances Li and Li are compared.
- In the following, the authors analyze the errors in the binarization process due to various global illumination effects and defocus, leading to systematic errors2.
- Such a situation can commonly arise due to long-range inter-reflections, when scenes are illuminated with lowfrequency patterns.
4. Patterns for Error Prevention
- Errors due to global illumination are systematic, scene-dependent errors that are hard to eliminate in 2Errors for the particular case of laser range scanning of translucent materials are analyzed in [7].
- The authors design patterns that modulate global illumination and prevent errors from happening at capture time itself.
- Thus, if the authors use low frequency patterns for short-range effects, the global component actually helps in correct decoding even when the direct component is low (Figure 3).
- For short-range effects, the authors want patterns with only low frequencies (high minimum stripe-widths).
4.1. Logical coding-decoding for long range effects
- The authors introduce the concept of logical coding and decoding to design patterns with only high frequencies.
- Thus, the authors can bypass explicitly computing the direct component.
- If the authors use the last Gray code pattern (stripe width of 2) as the base pattern, all the projected patterns have a maximum width of 2.
- Similarly, if the authors use the second-last pattern as the base-plane, they get the XOR-04 codes (Figure 5).
4.2. Maximizing the minimum stripe-widths for
- Short-range effects can severely blur the highfrequency base plane of the logical XOR codes.
- On the contrary, it is easy to generate codes with small maximum stripe-width (9), as compared to 512 for the conventional Gray codes, by performing a brute-force search min-SW (8) is given by Goddyn et al . [6].
- In comparison, conventional Gray codes have a minSW of 2.
- Kim et al . [11] used a variant of Gray codes with large min-SW called the antipodal Gray codes to mitigate errors due to defocus.
- Thus, these codes can be used in the presence of short-range effects as well.
4.3. Ensemble of codes for general scenes
- Global illumination in most real world scenes is not limited to either short or long range effects.
- For phase-shifting, the authors project 18 patterns (3 frequencies, 6 shifts for each frequency).
- On the other hand, scene points where only the two Gray codes agree correspond to translucent materials (sub-surface scattering).
- Conventional Gray codes and phase-shifting result in large errors.
- Only the logical codes (optimized for long-range interactions) are sufficient to achieve a nearly error-free reconstruction, instead of the full ensemble.
5. Error detection and correction
- The patterns presented in the previous section can successfully prevent a large fraction of errors.
- For highly challenging scenes, however, some errors might still be made (for example, see Figure 8).
- These codes can not handle the systematic errors made due to global illumination.
- Mark the camera pixels where no two codes agree as error pixels (Section 5), also known as Error detection.
- If the direct component is low (for example, in the presence of sub-surface scattering), this technique may not converge.
6. Limitations
- The authors methods assume a single dominant mode of light transport for every scene point.
- The authors thank Jay Thornton, Joseph Katz, John Barnwell and Haruhisa Okuda (Mitsubishi Electric Japan) for their help and support.
- A coaxial optical scanner for synchronous acquisition of 3d geometry and surface reflectance.
Did you find this useful? Give us your feedback
Citations
143 citations
[...]
132 citations
Cites background or methods from "Structured light 3D scanning in the..."
...lot of research activity in the last few years [7, 4, 8, 19, 5], for the first time, we present a technique which is fast, accurate and widely applicable....
[...]
...However, most current methods require several tens or hundreds of images [8, 6, 19, 4], require moving cameras or light sources [11, 5, 15], and are often limited to scenarios where only one global illumination effect is dominant [8, 19]....
[...]
...The phase error ∇φ = |φ(p) − φ(c)| results in systematic errors in the recovered shape [8, 7]....
[...]
130 citations
Cites background from "Structured light 3D scanning in the..."
...…liquids that have unknown dispersing media; or, (b) using structured lighting patterns [Mukaigawa et al. 2010], which provide only approximate isolation [Holroyd and Lawrence 2011; Gupta et al. 2011] and therefore induce errors in measured scattering parameters that are difficult to characterize....
[...]
...…of single scattering are potentially quite useful, but as discussed in the context of 3D surface reconstruction [Holroyd and Lawrence 2011; Gupta et al. 2011], they provide only approximate isolation, and there is currently no analysis of how this affects the accuracy of inferred…...
[...]
...Such lightingbased isolations of single scattering are potentially quite useful, but as discussed in the context of 3D surface reconstruction [Holroyd and Lawrence 2011; Gupta et al. 2011], they provide only approximate isolation, and there is currently no analysis of how this affects the accuracy of inferred scattering parameters....
[...]
...2010], which provide only approximate isolation [Holroyd and Lawrence 2011; Gupta et al. 2011] and therefore induce errors in measured scattering parameters that are difficult to characterize....
[...]
119 citations
95 citations
Cites background or methods from "Structured light 3D scanning in the..."
..., for 3D shape recovery [Gupta et al. 2011]) is not yet well understood....
[...]
...As such, use of the primal-dual coding framework in a non-coaxial arrangement (e.g., for 3D shape recovery [Gupta et al. 2011]) is not yet well understood....
[...]
References
782 citations
"Structured light 3D scanning in the..." refers background in this paper
...Significant progress has been made on both fronts (see the survey by Salvi et al [16]) as demonstrated by systems which can recover shapes at close to 1000 Hz....
[...]
484 citations
"Structured light 3D scanning in the..." refers background in this paper
...Recently, it was shown that the direct and global components of scene radiance could be efficiently separated [14] using high-frequency illumination patterns....
[...]
254 citations
"Structured light 3D scanning in the..." refers methods in this paper
...[13] presented an iterative approach for reconstructing shape of Lambertian objects in the presence of inter-reflections....
[...]
246 citations
"Structured light 3D scanning in the..." refers background in this paper
...[21] and at a depth resolution better than 30 microns [5]....
[...]
187 citations
Related Papers (5)
Frequently Asked Questions (11)
Q2. What is the way to design patterns with only high frequencies?
For short range effects, the authors draw on tools from the combinatorial maths literature to design patterns with large minimum stripe-widths.
Q3. What is the common way of detecting a scene?
In all these settings, scene points receive illumination indirectly in the form of inter-reflections, sub-surface or volumetric scattering.
Q4. What are the steps to correct the error?
It is important to note that for this error correction strategy to be effective, the error prevention and detection stages are critical.
Q5. What is the goal of this paper?
The goal of this paper is to build an end-to-end system for structured light scanning under a broad range of global illumination effects.
Q6. How can the authors find codes with large min-SW?
For conventional Gray codes, although short-range effects might result in incorrect binarization of the lower-order bits, the higherorder bits are decoded correctly.
Q7. Why did the early work in the field drive the research?
Since the early work in the field about 40 years ago [18, 12], research has been driven by two factors: reducing the acquisition time and increasing the depth resolution.
Q8. How many iterations do the authors need to reduce the residual errors?
Repeat steps 1 − 5 to progressively reduce the residual errors (Section 5).first iteration itself, the authors require only a small number of extra iterations (typically 1-2) even for challenging scenes.
Q9. What is the effect of low-pass filtering of the incident light?
In the context of structured light, these effects can severely blur the high-frequency patterns, making it hard to correctly binarize them.
Q10. What are the main issues that prevent them from being applicable broadly?
While these approaches have shown promise, there are three issues that prevent them from being applicable broadly: (a) the direct component estimation may fail due to strong inter-reflections (as with shiny metallic parts), (b) the residual direct component may be too low and noisy (as with translucent surfaces, milk and murky water), and (c) they require significantly higher number of images than traditional approaches, or rely on weak cues like polarization.
Q11. How is it possible to generate codes with large min-SW?
On the contrary, it is easy to generate codes with small maximum stripe-width (9), as compared to 512 for the conventional Gray codes, by performing a brute-force searchmin-SW (8) is given by Goddyn et al . [6]