LiDAR Detection Acceleration for Fast Obstacle and Drivable Area Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional laser radar systems face reliability issues due to low data calculation speed, which affects the accuracy and efficiency of obstacle detection and drivable area recognition in autonomous driving systems.
Innovation Solution
The method involves algorithm splitting and heterogeneous acceleration of feature algorithms to enable parallel processing of scanning data, specifically by splitting tasks such as edge extraction and classification storage, and utilizing heterogeneous computing devices to improve data operation speed and reduce CPU occupancy rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional laser radar sensing process is used, then measurement precision and recognition accuracy are improved, but data calculation speed deteriorates
Solution Approach 1:
The patent segments the feature algorithm into multiple sub-algorithms that can be processed in parallel. Specifically, the drivable area recognition algorithm is divided into grid image establishment, grid point labeling, and outer contour frame extraction sub-algorithms, which are then distributed to different processing units for concurrent execution, thereby improving data calculation speed while maintaining measurement precision
Solution Approach 2:
The patent introduces heterogeneous computing architecture as an additional processing dimension, utilizing both CPU and GPU resources simultaneously. The CPU handles control flow and complex logic while the GPU accelerates parallel computation tasks, creating a multi-dimensional processing framework that resolves the speed-accuracy tradeoff
2Device complexity
If conventional sequential processing is used, then algorithm complexity is reduced, but data operation speed deteriorates
Solution Approach 1:
The patent divides the sequential feature algorithm into independent parallel sub-algorithms including grid image establishment, grid point labeling, and outer contour frame extraction. Each sub-algorithm processes specific portions of the scanning data simultaneously, transforming a single-threaded complex algorithm into multiple simpler parallel processes that execute concurrently
Solution Approach 2:
The patent introduces a heterogeneous computing system as an intermediary layer between the laser radar sensor and the output results. This intermediary framework includes a processing unit that coordinates CPU and GPU resources, managing data flow and task distribution to achieve accelerated processing without increasing algorithmic complexity
Data Source
AI summary
A detection method and a device based on a laser radar, and a computer readable storage medium are disclosed. The detection method includes: obtaining scanning data of the laser radar; performing an algorithm splitting on a feature algorithm for detection based on the scanning data to obtain at least one sub-algorithm capable of parallel processing in the feature algorithm; and performing a heterogeneous acceleration for the at least one sub-algorithm to process the scanning data and to obtain a processing result; and obtaining a detected position of an obstacle and a detected drivable area based on the processing result.