Hybrid Voxel Feature Extractor for 3D Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D object detection methods face challenges in balancing voxel size for efficient computation and storage, as smaller voxels improve performance but increase inference time, while larger voxels fail to capture intricate features and accurately locate smaller objects.
Innovation Solution
The proposed solution involves a hybrid voxel feature extractor that performs hybrid scale voxelization and dynamic feature projection to generate a pseudo-image feature map, allowing for multi-class pyramid feature fusion and accurate 3D object box prediction, decoupling voxel feature extraction and pseudo-image feature map projection to enhance computational and storage efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a smaller voxel size is used, then detection performance and accuracy are improved, but inference time increases
Solution Approach 1:
The patent divides the feature extraction process into two independent segments: voxel feature extraction at multiple scales and pseudo-image feature map projection. This segmentation allows each module to operate independently with optimized parameters, enabling the system to use multiple small-scale voxels for accurate feature extraction while projecting to a manageable feature map size for efficient detection, thus resolving the contradiction between detection accuracy and inference time
Solution Approach 2:
The patent transitions from direct 3D voxel space to a 2D pseudo-image feature map space. By projecting multi-scale voxel features onto a 2D feature map, the system maintains the detailed spatial information from multiple voxel scales while operating in a dimensionally reduced space that enables faster computational processing and reduces inference time
2Productivity
If a larger voxel size is used, then computation and storage costs are reduced, but intricate features and accurate positions of smaller objects cannot be captured
Solution Approach 1:
The patent segments the feature extraction into multiple independent voxel scales (e.g., 0.05m, 0.1m, 0.2m). Each scale processes points independently and captures features at its appropriate resolution level. Small objects are captured by fine-grained voxels while large objects are efficiently processed by coarse voxels, achieving both computational efficiency and feature accuracy without requiring a single large voxel size
Solution Approach 2:
The patent applies different voxel scales to different spatial regions and object types. Fine-grained voxels (smaller size) are used where detailed feature extraction is needed for small objects, while coarse-grained voxels (larger size) are used for large objects and background regions. This local adaptation of voxel quality ensures computational efficiency is maintained while intricate features are captured where necessary
Data Source
AI summary
An object detection network includes: a hybrid voxel feature extractor configured to acquire a raw point cloud, extract a hybrid scale voxel feature from the raw point cloud, and project the hybrid scale voxel feature to generate a pseudo-image feature map; a backbone network configured to perform a hybrid voxel scale feature fusion by using the pseudo-image feature map to generate multi-class pyramid features; and a detection head configured to predict a three-dimensional object box of a corresponding class according to the multi-class pyramid features. The object detection network can effectively solve a problem that under a single voxel scale, inference time is longer if the voxel scale is smaller, and an intricate feature cannot be captured and a smaller object cannot be accurately located if the voxel scale is larger. Different classes of 3D objects can be detected quickly and accurately in a 3D scene.