LiDAR–Camera Likelihood Fusion for Drivable-Space Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing road drivable space detection methods face challenges such as instability due to complex weather changes and scene variations with single-sensor systems, and high computational complexity with multi-sensor fusion using neural networks.
Innovation Solution
A method involving multi-sensor and multi-cue fusion, utilizing laser radar for point cloud data processing and onboard camera images to generate likelihood maps, which are then fused using a Bayesian framework for accurate drivable space detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If single-sensor detection (onboard camera or laser radar) is used, then the device complexity is reduced, but the reliability and adaptability to complex weather and scene changes deteriorates
Solution Approach 1:
The patent combines laser radar and onboard camera into a unified detection system that processes both LiDAR point cloud data and visible light image data through multi-cue fusion algorithms, achieving improved reliability and weather adaptability while maintaining manageable system complexity through integrated processing architecture
2Measurement precision
If multi-sensor fusion detection using neural networks is used, then the detection accuracy and adaptability are improved, but the computational complexity increases
Solution Approach 1:
The patent replaces complex neural network-based multi-sensor fusion with a probability map fusion algorithm that processes laser radar and camera data through mathematical probability calculations, achieving high detection accuracy while significantly reducing computational complexity and processing time
Data Source
Figure 1
Figure 2
Figure 3~4b
AI summary
A method for detecting drivable space of a road includes obtaining radar point cloud data by a laser radar, and processing the radar point cloud data to obtain a radar likelihood map; capturing a visible image by an onboard camera, and performing the visible image with multi-cue processing to obtain a visible light likelihood map; fusing the radar likelihood map with the visible light likelihood map to obtain a road fusion map; and identifying based on the road fusion map to obtain a detection result for drivable space. The accuracy of the road detection is improved, which solves the technical problems of road detection by a single sensor, such as inability to adapt to complex varied weathers and scenes. Furthermore, the calculation in the present method is simpler than the traditional fusion detection (neural network detection).