Global Point Cloud Registration Using SHT for Faster Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing global registration algorithms face challenges in achieving high accuracy and speed due to non-convexity, leading to prolonged search times and local minimum convergence, making them unsuitable for applications requiring high precision and speed.
Innovation Solution
A global registration method utilizing spherical harmonic transformation (SHT) and iterative optimization, which includes voxelizing point clouds, aligning them in the frequency domain, and employing a branch and bound method with a new evaluation criterion based on hit rate and search time to efficiently narrow the search domain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional global registration algorithms are used, then the algorithm can handle non-convex optimization problems, but the search time becomes excessively long and the algorithm容易 falls into local minimum
Solution Approach 1:
The patent divides the search space into multiple subspaces based on the phase difference spectrum. By segmenting the search domain into regions around prominent peaks in the phase difference spectrum, the algorithm can systematically explore potential global optima while avoiding exhaustive search of the entire space, thus reducing search time while maintaining reliability in finding global optimal solutions.
Solution Approach 2:
The patent performs preliminary processing by computing the phase difference spectrum and identifying prominent peaks before the main optimization process. This preliminary action creates a roadmap of potential solution regions, allowing the subsequent optimization to focus computational resources on promising areas rather than randomly searching the entire parameter space, thereby reducing overall search time.
2Loss of time
If the search domain is narrowed quickly, then the search time is reduced, but the accuracy may not be sufficient
Solution Approach 1:
The patent applies different search strategies to different regions of the search space based on their characteristics. Regions around prominent peaks in the phase difference spectrum receive intensive search with higher accuracy requirements, while other regions are searched more broadly. This local differentiation allows the algorithm to spend computational resources where they are most needed for accuracy while maintaining overall speed.
Solution Approach 2:
The patent incorporates feedback mechanisms where the optimization process continuously evaluates the quality of solutions found and adjusts the search strategy accordingly. By monitoring the convergence behavior and solution quality, the algorithm can dynamically refine its search focus, ensuring that time is not wasted on low-quality regions while maintaining sufficient accuracy in the final result.
3Ease of manufacture
If feature point correspondence is used for global registration, then the algorithm can achieve registration, but the accuracy is reduced compared to original point cloud data
Solution Approach 1:
The patent introduces the phase difference spectrum as an intermediary representation that bridges the gap between computational efficiency and accuracy. Instead of directly working with feature point correspondences (which are easy to compute but less accurate) or raw point cloud data (which are accurate but computationally intensive), the phase difference spectrum serves as an intermediate that captures essential geometric information in a compressed, efficient representation, enabling both speed and accuracy.
Data Source
AI summary
A global registration method based on spherical harmonic transformation (SHT) and iterative optimization is provided. Two assumptions are provided: firstly, it is predefined that a minimum percentage of a correct matching pair in an input point cloud is represented as a limit on a quantity of outliers in the point cloud, and secondly, a distance threshold used to determine the correct matching pair is preset based on a scenario and represented as a limited distance of an outlier in the point cloud. In the algorithm provided, the point cloud first undergoes coarse registration to obtain a plurality of search domains, and the search domains are sorted based on an evaluation criterion. A branch and bound method is used to exclude an incorrect search domain and obtain a final registration result.