Hybrid Scale Voxelization for 3D Point Cloud Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for three-dimensional target detection face challenges in balancing voxel size for efficient computation and memory usage, as smaller voxel sizes improve performance for smaller objects but increase inference time, while larger voxel sizes fail to capture intricate features and accurate locations of smaller objects.
Innovation Solution
The method employs hybrid scale voxelization, generating point-wise hybrid scale voxel features and pseudo image feature maps using aggregated features and projection scale information, allowing for multiple voxel scales and dynamic projection scales to efficiently extract and fuse features, thereby improving computational and storage efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a smaller voxel size is used, then detection performance for smaller objects is improved, but inference time increases
Solution Approach 1:
The point cloud is segmented into multiple scale groups, where each group contains voxels of a specific scale. This segmentation allows the system to process different regions at appropriate scales, improving small object detection without uniformly increasing computation across the entire point cloud.
Solution Approach 2:
The patent introduces dynamic projection scales that can be adjusted based on the detection needs. By dynamically selecting projection scales for different scale groups, the system adapts the voxelization granularity to the specific detection task, balancing precision and inference time.
2Productivity
If a larger voxel size is used, then inference time is reduced, but feature capture accuracy for smaller objects deteriorates
Solution Approach 1:
Different regions of the point cloud are assigned different voxel scales based on their content. Regions containing small objects use finer voxel scales for accurate feature capture, while other regions use coarser scales to maintain inference speed. This local adaptation of quality resolves the contradiction between speed and accuracy.
Solution Approach 2:
The patent adds a scale dimension to the traditional single-scale voxelization by introducing multiple scale groups. This dimensional extension allows the system to represent the same spatial region at different granularities, enabling both fast processing and accurate feature capture simultaneously.
3Measurement precision
If multiple voxel scales are used, then feature extraction accuracy is improved, but computational complexity increases
Solution Approach 1:
Instead of processing all possible scale combinations, the patent applies partial action by selecting only the necessary scale groups for each detection task. This selective processing reduces computational complexity while maintaining the accuracy benefits of multi-scale feature extraction.
Solution Approach 2:
The multi-scale voxelization framework serves multiple functions: it enables accurate feature extraction, supports dynamic adaptation to different detection scenarios, and provides a unified structure for processing diverse object sizes. This multi-functionality justifies the increased computational complexity by delivering comprehensive detection capabilities.
Data Source
AI summary
A method for extracting point cloud feature includes: obtaining an original point cloud, conducting hybrid scale voxelization on the original point cloud; generating point-wise hybrid scale voxel features by feature encoding the point cloud subjected to the hybrid scale voxelization; and generating pseudo image feature maps using aggregated features and projection scale information. In this way, problems that at a single voxel scale, the inference time is longer when the voxel scale is smaller, and the larger voxel fails to capture intricate features and accurate location of smaller objects can be effectively overcome.