STEM与日常科技·英语精读30篇（5）
9 / 30
正在确认阅读权限…
LiDAR Point Cloud Degradation in Rain and Fog: Physics-Based Modeling of Signal Attenuation
激光雷达点云在雨雾中的退化：基于物理的信号衰减建模
-
Raindrops scatter 905 nm laser pulses through Mie scattering, reducing point cloud density by up to 73% at 25 mm/h intensity—degradation that scales nonlinearly with droplet size distribution, not just rainfall rate.
-
Fog attenuation follows exponential decay governed by Beer-Lambert law, but standard models underestimate backscatter from polydisperse aerosols, leading to false ‘free-space’ classifications in autonomous vehicle perception stacks.
-
Wavelength choice matters critically: 1550 nm lasers penetrate fog better than 905 nm but require more expensive InGaAs detectors and face stricter Class 1 eye-safety limits, constraining maximum pulse energy.
-
Real-time compensation requires dual-wavelength systems—measuring extinction at both bands to estimate Mie-to-Rayleigh scattering ratios—but adds optical complexity and calibration drift over thermal cycles.
-
Point cloud denoising algorithms trained on clear-weather data fail catastrophically in precipitation, misclassifying rain-induced speckle as static obstacles or occluding true pedestrians behind wet windshield streaks.
-
Sensor fusion mitigates but doesn’t eliminate risk: camera-based depth estimation degrades in low contrast, while radar lacks angular resolution to distinguish overlapping vehicles in dense traffic.
-
Automotive safety standards like ISO 21448 (SOTIF) now mandate rain/fog validation using calibrated nozzles that replicate droplet spectra—not just uniform water films.
-
Field data shows that lidar range reliability drops from 200 m in dry air to under 45 m in moderate fog (visibility 50 m), forcing fallback to conservative longitudinal control policies.
-
Physics-informed neural networks now embed Mie scattering cross-sections as hard constraints, improving generalization—but require GPU-accelerated ray tracing during inference.
-
Ultimately, robust autonomy demands not just better sensors but redefined operational design domains—acknowledging that some weather conditions remain fundamentally unperceivable with current photon budgets.