降雨条件下车载激光雷达感知局限性

Perception Limitation of LiDAR Sensor in Autonomous Vehicles Under Rainfall Condition

摘要:

针对激光雷达在降雨条件下的性能局限性问题，基于封闭场地中的降雨模拟设施，分别选取一款机械旋转雷达和混合固态激光雷达采集不同降雨条件下的点云数据，通过构建包括5项客观指标的指标集，对两类激光雷达受降雨条件影响的程度进行了对比分析。试验结果表明，降雨会在一定程度上减少点云的数量，降低点云的反射率，增加反射率信息熵，降低反射率信噪比；降雨的雾化现象会大幅恶化激光雷达的表现，在20m~40m内，对机械旋转雷达造成91.49%的平均点云数量削减和82.98%的平均反射率削减；而波长较长、非重复扫描的混合固态雷达在降雨下具有更好的穿透性，点云数量较为稳定；但混合固态雷达在降雨环境下的点云会包含更多反射率较低的雨滴回波，因此需要采用更有效地噪声滤除算法。

关键词:自动驾驶汽车;预期功能安全;激光雷达性能;降雨测试;性能局限

Abstract:

A study was made of the performance limitations of LiDAR under rainfall conditions. Based on the artificial rainfall facility in a proving ground， we collected point cloud data from a mechanical rotating LiDAR and a hybrid solid-state LiDAR separately. By constructing a metrics set including five objective indicators， the degree of influences of rainfall conditions on two kinds of LiDAR were compared and analyzed. The test results show that the point cloud number decreases， the reflectivity is weakened， but the information entropy of reflectivity increases.The signal-to-noise ratio reduces. The fogging phenomenon of rainfall significantly deteriorates the performance of the LiDAR. Within 20~40m， a 91.49% reduction of the average point cloud number and a 96.77% reduction of the average reflectivity of the mechanical rotating LiDAR are found. In comparison， the hybrid solid-state LiDAR with a longer wavelength and non-repetitive scanning has better penetrability under rainfall， resulting in a relatively stable number of point clouds. However， the point cloud of the hybrid solid-state LiDAR will contain more raindrop echoes with lower reflectivity， so a more effective noise filtering algorithm is necessary for application.

Key words:autonomous vehicle;safety of the intended functionality;LiDAR performance;rainfall experiment;performance limitations