FPGA Point Cloud Feature Extraction for Real-Time LiDAR Edge Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current point cloud feature extraction algorithms on edge computing processors achieve processing speeds lower than 9 Hz, which is not real-time capable and consumes excessive power, posing challenges for intelligent vehicles with high-speed LiDAR systems.
Innovation Solution
An energy-efficient point cloud feature extraction method is implemented on an FPGA, utilizing a laser angle guided projection method to organize unordered points into a matrix, followed by column-scanning and priority queue techniques to extract coarse-grained and fine-grained feature points efficiently, reducing computational complexity and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional feature extraction algorithms are used on edge computing processors, then the processing speed is lower than 9 Hz, but the power consumption is excessive
Solution Approach 1:
The patent replaces the software-based feature extraction algorithm running on edge computing processors with a hardware-based FPGA implementation. This substitution of computational platform fundamentally changes the processing architecture, enabling real-time operation at 20 Hz while significantly reducing power consumption through hardware optimization and parallel processing capabilities of the FPGA device.
2Speed
If the LiDAR scanning speed is increased to 20 Hz, then the real-time processing requirement is met, but the processing time of traditional algorithms cannot keep up
Solution Approach 1:
The patent implements a point cloud matrix organization structure that pre-arranges point cloud data in a grid-based format before feature extraction. This preliminary organization of data into matrices with defined row and column indices enables efficient random access and parallel processing during feature extraction, eliminating the time bottleneck that would otherwise prevent keeping up with 20 Hz LiDAR scanning rates.
Solution Approach 2:
The patent segments the point cloud data into a matrix structure divided into rows and columns, allowing independent processing of different matrix elements. This segmentation enables parallel computation across multiple processing units within the FPGA, significantly reducing the overall processing time required to analyze the entire point cloud at high scanning speeds.
3Measurement precision
If geometric feature points are calculated by computing local curvature of each point, then feature accuracy is improved, but the complexity increases and cannot efficiently process large quantities of point clouds
Solution Approach 1:
The patent applies local quality by organizing point cloud data into a matrix structure where each element represents a local region in the point cloud. Feature extraction is performed independently on each matrix element based on its local characteristics, allowing accurate curvature calculation only where needed while skipping redundant computations in uniform regions. This localized approach maintains feature accuracy while dramatically reducing overall computational complexity for large point clouds.
Data Source
AI summary
An energy-efficient point cloud feature extraction method based on a field-programmable gate array (FPGA) is mapped onto the FPGA for running. The energy-efficient point cloud feature extraction method based on the FPGA is applied to point cloud feature extraction in unmanned driving; or an intelligent robot. Compared with an existing technical solution, the energy-efficient point cloud feature extraction method based on the FPGA has following innovative points: a low-complexity projection method for organizing unordered and sparse point clouds, a high-parallel method for extracting a coarse-grained feature point, and a high-parallel method for selecting a fine-grained feature point.