3D Point Cloud Decimation for Real-Time Vehicle Navigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face high computational costs and processing delays when navigating using dense 3D point clouds from lidar sensors, as they require efficient processing of hundreds of thousands of points per second to extract static and dynamic objects in their environment.
Innovation Solution
The method involves generating a ground plane point cloud and an object point cloud through point cloud decimation using ground plane segmentation, applying a voxel filter, and selecting subsets of points to produce a representation of the ground plane, allowing for efficient navigation by distinguishing between ground and object points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dense 3D point clouds from lidar sensors are processed to extract static and dynamic objects, then navigation accuracy is improved, but processing time increases and computational cost increases
Solution Approach 1:
The patent segments the dense point cloud into ground plane points and object points by defining a ground plane representation and classifying points based on their relationship to this plane. This segmentation reduces the computational burden by separating navigation-relevant objects from the ground surface, enabling faster processing while maintaining navigation accuracy.
Solution Approach 2:
The patent extracts only the necessary object points from the dense point cloud after ground plane segmentation, discarding redundant ground points. This extraction process retains critical navigation information while eliminating unnecessary data, thereby reducing processing time without compromising navigation precision.
2Reliability
If dense 3D point clouds are processed to extract objects, then object detection completeness is improved, but computational cost increases
Solution Approach 1:
The patent segments the point cloud into ground and object components, allowing the system to focus computational resources on object detection while using the ground plane model to efficiently filter out non-object points. This maintains detection completeness by preserving all object points while reducing overall computational load.
Solution Approach 2:
The patent performs ground plane segmentation as a preliminary step before object detection. By pre-defining the ground plane representation and classifying points beforehand, the system reduces the complexity of subsequent object detection operations, maintaining completeness while lowering computational costs.
3Quantity of substance
If hundreds of thousands of points per second are processed, then point cloud density is maintained, but processing speed decreases
Solution Approach 1:
The patent extracts and retains only the essential object points from the dense point cloud after ground plane segmentation, removing redundant ground points. This extraction maintains sufficient point cloud density for accurate object representation while dramatically increasing processing speed by reducing the total number of points that must be processed.
Solution Approach 2:
The patent applies different processing quality levels to different parts of the point cloud: ground points are processed to define the plane representation, while object points are retained with higher fidelity. This local quality approach maintains necessary density where needed (object regions) while reducing density elsewhere (ground regions), improving overall processing speed.
Data Source
AI summary
A method for navigation of a vehicle using point cloud decimation includes generating, by a processor circuit, a ground plane point cloud and an object point cloud. Generating the ground plane point cloud and the object point cloud includes performing point cloud decimation of a 3-D point cloud using ground plane segmentation. The further includes navigating, by the processor circuit, the vehicle using at least the object point cloud.