FPGA KNN Accelerator for Point Cloud Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing K-nearest neighbors search algorithms for large-scale point cloud maps face challenges such as redundant search areas and slow data transmission, leading to inefficiencies and long construction times, which are unsuitable for real-time applications in autonomous vehicles.
Innovation Solution
The proposed solution is a fast and energy-efficient K-nearest neighbors search accelerator based on the NSVS framework, which constructs a DSVS search structure and implements the search algorithm on an FPGA. This framework reduces redundant search areas through nearest sub-voxel selection and optimizes data transmission based on data reuse rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If DSVS-based hardware accelerator is used to quickly construct large-scale point cloud map, then construction speed is improved, but data transmission time increases and search area becomes redundant
Solution Approach 1:
The patent segments the point cloud map into multiple tiles, where each tile is independently processed and stored. This segmentation allows the system to load only the necessary tiles into memory during search operations, significantly reducing data transmission time while maintaining construction speed. The tile-based structure enables selective access to relevant portions of the large-scale point cloud map without transmitting the entire dataset.
2Productivity
If DSVS-based hardware accelerator is used to quickly construct large-scale point cloud map, then construction speed is improved, but search area becomes redundant
Solution Approach 1:
The patent implements adaptive search area adjustment based on local point cloud density and query characteristics. The system dynamically determines the optimal search radius and scope for each query operation, expanding the search area only when necessary to find sufficient neighbors. This local quality approach eliminates redundant search in sparse regions while ensuring adequate search coverage in dense regions, thereby reducing overall search area without compromising search quality.
3Measurement precision
If conventional K-nearest neighbors algorithm is used for large-scale point cloud map, then search accuracy is maintained, but search time becomes excessively long
Solution Approach 1:
The patent performs preliminary organization of the point cloud map into a tile-based hierarchical structure during the construction phase. This preliminary action creates an optimized data layout that enables fast retrieval during search operations. By pre-processing the data into manageable tiles with appropriate indexing, the system can quickly locate and access relevant point clouds without performing exhaustive searches, thus maintaining search accuracy while dramatically reducing search time for large-scale maps.
Data Source
AI summary
A fast and energy-efficient K-nearest neighbors search accelerator for a large-scale point cloud is provided. A nearest sub-voxel-selection (NSVS) framework that performs search based on a double-segmentation-voxel-structure (DSVS) search structure is constructed, and a K-nearest neighbors search algorithm for a large-scale point cloud map is implemented on a field programmable gate array (FPGA). The K-nearest neighbors search accelerator is configured for constructing the DSVS search structure, and searching for K-nearest neighbors based on the DSVS search structure. An experimental result on a KITTI dataset shows that the K-nearest neighbors search accelerator has a search speed 9.1 times faster than a state-of-the-art FPGA implementation. In addition, the K-nearest neighbors search accelerator also achieves an optimal energy efficiency, and the optimal energy efficiency is 11.5 times and 13.5 times higher than state-of-the-art FPGA and GPU implementations respectively.