K-ary Tiled Patterns for Structured Light Depth Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for capturing three-dimensional scene depth information using structured light patterns face challenges such as large window sizes due to binary codes, which increase the impact of local errors and occlusion, and are not robust to noise and surface reflectivity variations, while also being computationally expensive and limited by spatial resolution.
Innovation Solution
A two-dimensional pattern is developed using tiled Toroidal Binary Perfect Submaps (TBPS) or Toroidal R-ary Perfect Submaps (TRPS) with periodicity, allowing for smaller window sizes and robust recovery of codes, even in non-disjoint regions, by projecting a sequence of phase-shifted images to account for spatially variable intensity variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If binary codes (de Bruin arrays or M-arrays) are used for structured light patterns, then the pattern is easy to construct, but the window size becomes large which increases the impact of local errors and occlusion
Solution Approach 1:
The patent transitions from binary codes (K=2) to K-ary codes where K>2, changing the parameter of code base. This allows each window code to represent more information with fewer points, reducing the window size from m≧log2(M) to m≧2logK(M) points while maintaining the ability to uniquely identify locations and reducing the impact of occlusion and local errors
2Ease of manufacture
If binary codes are used, then the pattern construction is simple, but the pattern is not robust to noise and surface reflectivity variations
Solution Approach 1:
The patent uses K-ary codes with K>2 instead of binary codes, changing the code base parameter. This provides better robustness to noise and surface reflectivity variations because the larger alphabet size allows for greater separation between valid codes in the Hamming space, making the system more tolerant to measurement errors and intensity variations
3Loss of information
If all or almost all possible window codes are used exactly once, then the pattern provides dense correspondences, but the pattern is not robust to measurement errors
Solution Approach 1:
The patent extracts and uses only a sparse subset of all possible K-ary window codes rather than using all or almost all possible codes. This sparse usage allows for better error robustness while still providing sufficient correspondences for accurate depth measurement, as the selected codes are chosen to maximize separation in Hamming space
Solution Approach 2:
The patent pre-selects window codes with appropriate Hamming distance properties before generating the pattern. This beforehand selection cushions against measurement errors by ensuring that valid codes are sufficiently separated in code space, allowing error detection and correction capabilities to be built into the pattern structure itself
4Reliability
If a sparse subset of window codes is used, then error detection and correction is improved, but the pattern requires careful code selection
Solution Approach 1:
The patent uses systematic methods for selecting sparse subsets of K-ary codes, such as using codes with guaranteed minimum Hamming distance properties or using algebraic structures like Reed-Solomon codes. This systematic approach manages the complexity of code selection while maintaining strong error detection and correction capabilities
Data Source
AI summary
A two-dimensional pattern comprises a plurality of R-planes each comprising a tiling of a corresponding R-ary block, being a block of radix R integer values, where for each dimension of the pattern, the least common multiple of the sizes of the tiled blocks in that dimension is greater than the size of the tiling that dimension, and any sub-block of a size less than the tiled blocks occurs on a regular grid with the same periodicity as the tiled block for that R-plane. The pattern may be used in determining a position of a location captured in an image by projecting the pattern onto a scene. An image is captured. The method determines from the captured image a sub-block associated with the location and constructs, a unique integer value for each R-plane. The unique integer values from each R-plane are used to determine the location in the image.