Hierarchical Data Structure for Sensor Point Cloud Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for aligning sensor measurement data with reference data, such as LIDAR, TOF camera, and RADAR systems, face high complexity and computational demands, making them unsuitable for real-time applications with limited processing resources.
Innovation Solution
A method that uses a hierarchical data structure with cells organized in N levels to iteratively fit data points by determining cell features and level-transforms, reducing the need for intensive correspondence mapping and allowing for efficient processing of sensor data from various sources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If iterative closest point (ICP) method is used for sensor data registration, then alignment accuracy is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent divides the data sets into multiple subsets based on spatial regions or feature groups. Instead of performing correspondence matching on the entire data set, the method processes smaller subsets independently, determining partial transforms for each subset and combining them to achieve the final alignment. This segmentation reduces the computational complexity of the correspondence problem while maintaining alignment accuracy.
2Manufacturing precision
If traditional ICP correspondence mapping is performed, then fitting accuracy is improved, but processing speed decreases
Solution Approach 1:
The patent performs preliminary organization of data into subsets and pre-computation of certain transformation parameters before the main fitting process. By preparing data structures and identifying potential correspondences in advance, the method reduces the computational burden during the actual iterative fitting process, thereby improving processing speed without sacrificing fitting accuracy.
3Measurement precision
If complete correspondence mapping between data points is performed, then registration precision is improved, but resource consumption increases
Solution Approach 1:
The patent performs correspondence mapping only on selected subsets of data points rather than attempting to map every single point. By identifying and processing only the most informative or critical subsets (such as feature-rich regions or boundary points), the method achieves sufficient registration precision while significantly reducing the computational resources required compared to complete correspondence mapping.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
There is disclosed a method for fitting a first set of data points to a second set of data points. The method comprises obtaining (S1) first and second sets of data points; defining (S2) a common data structure comprising cells organized hierarchically in N levels, wherein each cell is associated with a respective cell range of data point values; associating (S3) data points in the first set and in the second set with cells at an initialization level of the common data structure based on cell ranges and on data point values; and, for each level j, starting from the initialization level; determining (S4) at least one cell feature for one or more cells and for each data set based on data points associated with respective cells; determining (S5) a level-transform Tj for level j based on differences between respective cell features determined at level j, to fit the second data set to the first data set at level j by reducing differences between cell features; associating (S6) data points in the first set, and data points transformed by level-transform Tj in the second set, with cells at a next level j+1 of the common data structure based on cell ranges at level j+1; until a stopping criterion is fulfilled, and; fitting (S7) the first set of data points to the second set of data points by accumulating the determined level-transforms Tj.