基于KITTI生成的恶劣天气下点云数据

基于公开数据集KITTI中的点云数据生成的恶劣天气下的点云数据。这些原始点云文件以.bin格式存储，包含了每个点的三维坐标信息（x,y,z）以及其他信息如反射强度，使用Python中的numpy库加载这些数据，通过读取每帧对应的点云文件，将其转化为点云的三维坐标数据。强光照天气通常会使得激光雷达的反射信号增强，特别是在阳光直射下的场景中。为了模拟这一现象，我们对点云的反射强度进行处理，将反射强度值乘以一个系数来实现，如设置系数大于1则反射强度加倍，反之亦然。此外，点云的反射信号也会受到不同角度的光照影响，我们选择在点云中应用一定的角度修正，对于不同的点，基于其法向量和光源方向的夹角，可以模拟出反射强度的变化。对于弱光照条件，我们沿用强光照的方法，将系数设置为小于1，通过这种方法可以将反射强度降低。此外，在弱光照条件下，由于激光雷达的信号质量下降，点云数据中的噪声通常会增加，我们选择在点云中添加一些随机的高斯噪声来模拟这一现象。

Point cloud data under adverse weather conditions is generated using the point cloud data from the public KITTI dataset. These raw point cloud files are stored in .bin format, containing the 3D coordinate information (x, y, z) of each point as well as auxiliary attributes such as reflection intensity. We utilize the numpy library in Python to load these datasets: by reading the point cloud files corresponding to each frame, we convert them into the 3D coordinate data of the point cloud. Intense sunlight typically enhances lidar reflection signals, particularly in scenes under direct solar irradiation. To simulate this effect, we adjust the reflection intensity of the point cloud by multiplying the original intensity values by a scaling factor: setting the factor greater than 1 amplifies the reflection intensity, while setting it below 1 reduces it. Additionally, the reflection signals of the point cloud are influenced by the relative angle between illumination and the scene. We apply angle correction to the point cloud: for each individual point, the variation in reflection intensity can be simulated based on the angle between its normal vector and the light source direction. For low-light conditions, we follow the same approach as for intense sunlight, using a scaling factor less than 1 to reduce the reflection intensity. Furthermore, under low-light conditions, the noise in point cloud data typically increases due to degraded lidar signal quality. To simulate this phenomenon, we add random Gaussian noise to the point cloud data.

搜集汇总

背景与挑战

背景概述

该数据集是基于KITTI公开数据集生成的模拟恶劣天气条件下的点云数据，通过调整反射强度系数和添加高斯噪声来模拟强光照、弱光照等天气影响，旨在提升车联网和自动驾驶系统在复杂环境下的感知能力。数据以.bin格式存储，包含三维坐标和反射强度信息，总数据量为13.27GB，共7482个文件，适用于相关算法研究和测试。