3D Point-Cloud Search Using Dimensional Transform and Voxel Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing point-cloud registration methods, such as Iterative Closest Point (ICP) and Generalized Iterative Closest Point (GICP), are computationally inefficient due to expensive tree traversals and re-balancing, making real-time processing impractical for desired precision in applications like computer vision, SLAM, and autonomous driving.
Innovation Solution
A dimensional transform (DST) maps a 3D space to a one-dimensional space with an implicit recursive hierarchical structure, reducing neighborhood searches to one-dimensional searches, and incorporates black-white arrays (BWA) for efficient point-cloud voxelization and voxel map operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If tree-based approaches (K-d trees) are used for 3D space searching in point-cloud registration, then search functionality is provided, but computational cost becomes expensive due to tree traversals and re-balancing
Solution Approach 1:
The patent transforms the 3D space search problem into a 1D space search problem by applying a dimensional transform (DST) that maps three-dimensional coordinates to one-dimensional keys. This dimensionality reduction eliminates the need for complex 3D tree traversals while preserving spatial relationships, thereby resolving the contradiction between search precision and processing speed.
Solution Approach 2:
The patent replaces the mechanical tree-based search structure with a direct 1D array indexing system. Instead of traversing and rebalancing tree structures in 3D space, the system uses simple 1D array lookups after dimensional transformation, substituting a complex mechanical search system with a simpler, faster computational approach.
2Productivity
If point-cloud frames are compressed by sampling to reduce cardinality, then processing costs decrease, but accuracy may significantly decrease
Solution Approach 1:
The patent applies dimensional transform to compressed point-cloud data, allowing efficient processing of reduced datasets while maintaining accuracy through the transformed space representation. The DST-based search in 1D space preserves spatial relationships even when point density is reduced, enabling accurate registration with fewer points.
3Measurement precision
If iterative searching is performed in 3D spaces for ICP and GICP methods, then closest points are found for alignment, but performance is dominated by the cost of such searches
Solution Approach 1:
The patent transforms iterative 3D space searching into a single-pass 1D space search using dimensional transform. The DST mapping converts the iterative closest point search into a direct 1D array lookup operation, eliminating multiple iterative search cycles while maintaining the ability to find closest points for accurate alignment.
Data Source
AI summary
The current document is directed to systems and methods that incorporate a dimensional transform that maps a 3D space to a one-dimensional space with an implicit recursive hierarchical structure that preserves 3D neighborhoods. A regional or neighborhood search for points in a 3D subspace is reduced, by the dimensional transform, to one or more searches in the transformed 1D space. The search is performed by either recursive decomposition of the 3D space indexed by the transform into subspaces, exploiting the transformed space structure, or by direct indexing into the region of interest. Performance of the control systems and methods is further enhanced by incorporating one or more black-white arrays, which support-searches over one-dimensional spaces as well as dynamic 3D spaces. Dimensional-transform-based methods also provide for point-cloud voxelization and voxel map operations.