Voxel Ground Plane Segmentation for Static and Dynamic Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Segmenting three-dimensional data, particularly for autonomous vehicles using LIDAR systems, is challenging due to the complexity of distinguishing between static and dynamic objects and accurately identifying a ground plane within voxel spaces.
Innovation Solution
The method involves representing LIDAR data in a voxel space, determining locally flat voxels to identify the ground plane, using clustering techniques to segment objects, and employing ray casting to differentiate between static and dynamic objects, thereby facilitating efficient object tracking and navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image segmentation methods are applied to three-dimensional LIDAR data, then object identification can be achieved, but processing complexity and computational requirements increase significantly
Solution Approach 1:
The patent applies segmentation by dividing the three-dimensional LIDAR data into discrete voxels (volume elements) that can be processed independently. Each voxel represents a small volumetric unit in the point cloud data, allowing the system to segment and analyze objects through voxel-based operations rather than processing the entire point cloud as a single complex structure. This voxelization enables efficient ground plane detection and object segmentation by treating the data as a grid of manageable units.
2Measurement precision
If clustering techniques are used to segment objects in voxel space, then object segmentation accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent extracts and removes ground plane voxels from the voxel space before performing object segmentation clustering. By identifying and separating ground-related voxels (which form the baseline surface) from other voxels, the system reduces the total number of voxels that need to be processed during clustering operations. This extraction step simplifies the clustering problem by eliminating the ground plane as a confounding factor, thereby reducing processing time while maintaining segmentation accuracy.
3Measurement precision
If ray casting is employed to differentiate static and dynamic objects, then object classification accuracy improves, but computational load increases
Solution Approach 1:
The patent performs preliminary action by pre-processing the LIDAR data into voxel space and identifying ground plane voxels before executing ray casting operations. This preliminary voxelization and ground separation creates an optimized data structure that reduces the number of rays needed for classification and minimizes the computational scope of subsequent ray casting operations. By preparing the data in advance with ground removal, the system reduces the computational load during the actual object classification phase.
4Reliability
If the ground plane is accurately identified in voxel space, then navigation reliability improves, but the complexity of ground plane detection increases
Solution Approach 1:
The patent introduces an intermediary approach by using voxel-based representation as a mediator between raw LIDAR point cloud data and ground plane detection algorithms. Instead of directly analyzing complex three-dimensional point clouds, the system transforms data into a voxel grid where ground plane detection can be performed through simpler operations on discrete volumetric units. This intermediary voxel representation simplifies the detection process while maintaining navigation reliability.
Data Source
AI summary
Systems, methods, and apparatuses described herein are directed to performing segmentation on voxels representing three-dimensional data to identify static and dynamic objects. LIDAR data may be captured by a perception system for an autonomous vehicle and represented in a voxel space. Operations may include determining a drivable surface by parsing individual voxels to determine an orientation of a surface normal of a planar approximation of the voxelized data relative to a reference direction. Clustering techniques can be used to grow a ground plane including a plurality of locally flat voxels. Ground plane data can be set aside from the voxel space, and the remaining voxels can be clustered to determine objects. Voxel data can be analyzed over time to determine dynamic objects. Segmentation information associated with ground voxels, static object, and dynamic objects can be provided to a tracker and/or planner in conjunction with operating the autonomous vehicle.