LidarSIM 多 LiDAR 感知仿真与评价平台

LidarSIM — Multi-LiDAR Simulation & Evaluation

LidarSIM 面向动态交通场景,基于 GPU 并行 ray-casting 仿真多路 路侧 / 车载 LiDAR 点云,提供物理真实感的扫描时序运动伪影 (rolling shutter)、强度衰减(反射率 × 入射角 × 大气消光)、 雨/雪天气杂波、漏检模型等功能,内置完整的协同感知评估链路。

LidarSIM targets dynamic traffic scenarios. It simulates multi-roadside / on-board LiDAR point clouds via GPU-parallel ray-casting with physically accurate scanning-order motion artefacts (rolling shutter), intensity attenuation (reflectance × incidence × atmospheric loss), rain/snow clutter and dropout — plus a built-in cooperative-perception evaluation pipeline.

输入

Input

  • • 轨迹文件(Vissim / SUMO 格式)
  • • 道路环境点云(PLY / H5 含语义)
  • • 车辆三维模型(STL / PLY / OBJ)
  • • Vehicle trajectories (Vissim / SUMO formats)
  • • Environment point cloud (PLY / H5 with semantics)
  • • Vehicle 3D mesh library (STL / PLY / OBJ)

核心能力

Capabilities

GPU 并行 ray-casting 5 种 LiDAR 预设 物理强度模型 雨/雪天气 运动伪影 L-shape 评估 概率占用网格 OpenCOOD 导出
GPU ray-casting 5 LiDAR presets Physical intensity Rain / snow Motion artefacts L-shape evaluation Probabilistic occupancy grid OpenCOOD export

功能模块

Functional Modules

① LiDAR 仿真

① LiDAR Simulation

多类型 LiDAR 预设 · 雨雪天气 · 车路协同 · Ego 视角

Multi-type presets · weather · V2X · ego view

多类型路侧 LiDAR 仿真

Multi-type roadside LiDAR

雨雪天气仿真

Rain / snow weather simulation

车路协同感知 (V2X)

Vehicle–road cooperative perception (V2X)

Ego vehicle sim

单车 Ego 视角仿真

Ego-vehicle simulation

多 LiDAR 差异化布设展示

Multi-LiDAR placement comparison

② 感知性能评价

② Perception Evaluation

L-shape 检测 · IoU / Dev / Comp · 分类型评估 · CSV 导出

L-shape · IoU / Dev / Comp · per-vType breakdown · CSV

评价结果输出演示

Evaluation workflow

IoU/Dev sequence

IoU / Dev / Comp 数据展示(CSV)

IoU / Dev / Comp time series (CSV)

③ 概率占用网格 (POG)

③ Probabilistic Occupancy Grid

逐帧占用平均 · 观测掩码 · BEV / 3D 可视化 · NPZ / MAT 导出

Per-frame occupancy · observation mask · BEV/3D · NPZ/MAT export

POG 动态构建过程

POG construction over time

POG result

POG 结果可视化

POG visualization (BEV)

标注数据导出 (OpenCOOD)

Annotation Export for OpenCOOD

为支持协同感知模型训练,指定起止帧后,LidarSIM 支持导出 OpenCOOD 框架(OPV2V 格式)下的标注数据: 每帧生成协同感知点云与 GT BBox,自动按场景编号归档, 可直接用于多车辆协同感知模型训练。

To support cooperative-perception training, LidarSIM exports OpenCOOD-format (OPV2V) annotations given a frame range. Each frame produces cooperative point clouds and GT BBoxes, organised per scene and ready to plug into multi-vehicle cooperative-perception pipelines.

scene_00x / 00x / xxxxx.pcd OpenCOOD yaml GT BBox 标注