Point Cloud Registration Using Surface-Normal NDT
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The Normal Distribution Transformation (NDT) method for registering point clouds from laser scanners is inadequate in modeling points with sufficient precision, particularly for inner and outer walls and corners, and struggles to distinguish between physically meaningful and impossible registrations, leading to errors in mapping.
Innovation Solution
The Surface Normal Normal Distribution Transformation (ONDT) method, which determines surface normals at points, groups them based on these normals, and calculates multiple normal distributions within cells, providing a more accurate and physically meaningful representation for registration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the Normal Distribution Transformation (NDT) method is used to represent point clouds in cells, then computational efficiency is improved, but modeling precision for inner and outer walls and corners deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the point cloud representation into multiple groups based on surface normal orientations within each cell. Instead of treating all points in a cell uniformly, the method segments them according to their normal vectors, creating distinct groups for different surface orientations (e.g., inner walls, outer walls, corners). This segmentation allows each group to be modeled with appropriate precision while maintaining overall computational efficiency.
Solution Approach 2:
The patent implements local quality by applying different normal distribution models to different regions within cells based on their surface characteristics. Points with similar surface normals are grouped together and modeled with localized normal distributions, allowing the representation to adapt to local geometric features such as corners and walls. This ensures high modeling precision for critical features while maintaining efficiency through selective detailed representation.
2Speed
If the NDT approach is used for point cloud registration, then registration speed is improved, but the ability to distinguish physically meaningful registrations from impossible ones deteriorates
Solution Approach 1:
The patent introduces surface normal information as an intermediary variable that mediates between the point cloud data and the registration process. By incorporating normal vectors as an additional dimension in the representation, the method provides physical constraints that help distinguish meaningful registrations from impossible ones. This intermediary information guides the registration to physically plausible solutions while maintaining computational efficiency through the structured normal-based grouping.
3Measurement precision
If surface normals are determined and points are grouped based on normals, then representation precision is improved, but device complexity increases
Solution Approach 1:
The patent applies parameter changes by utilizing surface normal vectors as additional parameters for point classification and grouping. Instead of relying solely on spatial coordinates, the method incorporates normal direction parameters to define groups within cells. This parameter expansion improves representation precision by capturing surface orientation information, while the systematic approach to normal calculation and grouping manages the increased complexity through efficient algorithms.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
Revealbart is a method for reducing ambiguities in the representation and registration of point clouds. The point clouds are generated by at least one laser scanner or extracted from a navigation map. A first point cloud and a second point cloud are received as input.The procedure comprises the following steps: a) determining surface normals (24, 25, 26, 27) at the points of the first point cloud (7) and at the points of the second point cloud, b) determining a representation of the first point cloud, comprising the following substeps: i) subdividing the first point cloud (7) into cells (9, 10, 11, 12), ii) grouping the points of each cell (9, 10, 11, 12) based on the surface normals (24, 25, 26, 27) at the points of the first point cloud (7), iii) determining the normal distribution for the points of each group in each cell, and c) registering the second point cloud to the first point cloud (7) based on the representation. The core of the procedure is the introduction of an additional variable in the form of surface normals into the Normal-Distribution Transformation (NDT) approach, specifically through step a) and substep ii).This is an extension of the standard NDT method, resulting in a more precise representation of a point cloud, particularly within a cell, and a more physically meaningful registration of point clouds. The method is especially advantageous for laser data-based mapping of autonomous work machines, minimizing the erroneous detection of interior and exterior walls as well as corners without compromising localization efficiency.