摘要
自动驾驶仿真数据集在自动驾驶算法迭代等方面有着天然的优势,而大部分的仿真数据集都是基于CARLA生成的,CARLA是一个广泛应用的自动驾驶仿真平台,为自动驾驶感知、决策等方面都起到了强大的赋能作用。然而,CARLA对激光雷达模型的构建有着严重缺陷,其只考虑到了点云的3维空间信息而忽略了反射率的信息,基于BRDF模型对CARLA中的激光雷达模型进行了重建,使其能够返回带有较为准确反射率的点云,并搭建了一个自动化的多模态仿真数据集生成框架,从而生成多模态的自动驾驶仿真数据集。
Automatic driving simulation data sets have natural advantages in automatic driving algorithm iteration,and most of the simulation data sets are generated based on CARLA.CARLA is a widely used automatic driving simulation platform,which plays a powerful enabling role in automatic driving perception,decision making and other aspects.However,CARLA has serious defects in the construction of the LIDAR model,which only considers the spatial information of the point cloud and ignores the reflectivity information.The LIDAR model is reconstructed in CARLA based on the BRDF model,so that it can return a point cloud with more accurate reflectivity.An automatic multi-modal simulation data set generation framework has also been built to generate multi-modal automatic driving simulation data set in this work.
作者
王宇唯
黄宏成
Wang Yuwei;Huang Hongcheng(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240)
出处
《传动技术》
2023年第4期3-6,共4页
Drive System Technique
