LiDAR Point Cloud Clustering via Geometric Sub-Grid Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LiDAR-based object detection systems face issues with information loss and mis-clustering due to the grid-based clustering method, which can group separate objects into a single cluster.
Innovation Solution
A LiDAR-based object detection method that improves clustering by selecting a to-be-divided cluster, identifying division points based on geometrical features, and dividing the cluster using a representative point determined by these division points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If grid-based clustering is used to process LiDAR points, then real-time processing capability is improved, but information loss and mis-clustering occur
Solution Approach 1:
The patent applies segmentation by dividing the point cloud data into multiple grids, then further segmenting each grid into sub-grids for more detailed processing. This hierarchical segmentation allows real-time processing while preserving more information by examining smaller regions more closely, thereby reducing mis-clustering of separate objects.
Solution Approach 2:
The patent implements local quality by applying different processing resolutions to different regions. Important regions (where objects are detected) receive finer sub-grid analysis, while less important areas use coarser grid processing. This approach maintains real-time performance while improving detection accuracy in critical areas and reducing information loss.
2Measurement precision
If grid resolution is increased to reduce mis-clustering, then object detection precision is improved, but computational cost increases
Solution Approach 1:
The patent segments each grid into multiple sub-grids, enabling high-resolution analysis only where needed. This segmentation allows the system to achieve high detection precision in regions of interest while maintaining lower resolution in other areas, thus controlling computational cost.
Solution Approach 2:
The patent applies local quality by using finer sub-grid resolution specifically in grids containing detected objects or potential objects, while using coarser resolution elsewhere. This selective refinement improves object detection precision without uniformly increasing computational cost across the entire point cloud.
Solution Approach 3:
The patent applies partial action by performing detailed sub-grid analysis only on selected grids that contain objects or potential objects, rather than processing all grids at high resolution. This approach achieves high detection precision for relevant objects while avoiding the computational overhead of processing the entire scene at maximum resolution.
Data Source
AI summary
A LiDAR-based object detection method includes clustering a point cloud acquired from LiDAR, selecting a to-be-divided cluster among clusters generated in the clustering, and selecting division points according to a geometrical feature formed with adjacent points from among points belonging to the to-be-divided cluster, and dividing the to-be-divided cluster based on a representative point determined by at least some of the division points.