3D Point Cloud Alignment Using Reference Area Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for aligning three-dimensional point cloud data from multiple sensors face accuracy issues due to the presence of similar objects in the environment, leading to errors in feature point association and potential inversion of data sets.
Innovation Solution
A three-dimensional point cloud aligning device that allows users to designate a reference area within three-dimensional images displayed on a screen, using a machine learning model to suggest candidates and adjust frames for alignment, and computes a homogeneous transformation matrix to align partial data sets accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If feature point association is used for aligning point cloud data, then alignment can be performed automatically, but alignment accuracy deteriorates when similar objects are present in the environment
Solution Approach 1:
The patent divides the point cloud data into multiple regions and performs alignment separately for each region. By segmenting the data space and applying region-specific transformation parameters, the system avoids incorrect feature point associations caused by similar objects in different locations, thereby maintaining both automation and accuracy.
Solution Approach 2:
The patent applies different alignment strategies and transformation parameters to different regions of the point cloud data. Each region is processed with locally optimized parameters, allowing the system to adapt to local characteristics and avoid the pitfalls of global feature point matching when similar objects are present.
2Device complexity
If feature point association is used for alignment, then the process can be simplified, but alignment accuracy deteriorates due to errors in associating feature points
Solution Approach 1:
The patent segments the point cloud data into multiple regions and performs alignment separately for each region. This segmentation approach maintains process simplicity while improving accuracy by preventing incorrect feature point associations between similar objects located in different regions.
Solution Approach 2:
The patent introduces region division as an intermediary step between data acquisition and alignment. This intermediary mechanism organizes the data spatially before alignment, making the process more robust without significantly increasing complexity, as the region-based approach provides a clear framework for subsequent alignment operations.
3Ease of operation
If global alignment is performed on entire point cloud data, then processing is straightforward, but alignment accuracy deteriorates when similar objects are present
Solution Approach 1:
The patent divides the global point cloud data into multiple regions and performs alignment operations on each region separately. This segmented approach maintains operational simplicity by using consistent alignment procedures while improving accuracy through region-specific processing that avoids confusion from similar objects.
Solution Approach 2:
The patent applies alignment operations to partial regions of the point cloud data rather than attempting to align the entire dataset globally. By focusing alignment efforts on specific regions, the system achieves better local accuracy without the errors that arise from global feature point matching when similar objects are present.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An acquisition unit (32) acquires plural sets of three-dimensional point cloud data that are respective sensing results from plural sensors that detect three-dimensional positions of points in a peripheral environment. An accepting unit (34) displays each of three-dimensional images representing each of the plurality of sets of three-dimensional point cloud data on a screen. The accepting unit (34) accepts a designation in each of the three-dimensional images of an area that serves as a reference for alignment between the plural sets of three-dimensional point cloud data. An adjusting unit aligns partial sets of three-dimensional point cloud data with one another, the partial sets of three-dimensional point cloud data corresponding to the area in each of the plurality of sets of three-dimensional point cloud data.