数据收集代码-V5

[wmn7]

版本五

• 时间：2025.07.29
• 代码收集版本汇总记录

总结

• 包含 12 个真实路网
• 每个路网包含 5 种车流，模拟一天中的不同时段和任务
• 包含不同的特殊事件，每一个路网存在

内容介绍

• 每一次仿真里面，事故可以发生多次，每次时间 30-60s
• 特殊车辆不要右转，事故和特殊车辆不要同时出现
• 查看图片和文本是否可以对应，可以生成文本来描述图片
• 渲染的场景, 支持白天和黑夜(路灯)
• 韩国路网进行更新，只保留研究区域（研究的单路口）
• 所有车流文件更新新的车辆类型，可以模拟事故
• 修改 config 文件，每个 env 的 config 文件独立保存，环境和事故也独立
• 事故发生的位置可以增加 0.5
• 需要安装 hydra 库

pip install hydra-core --upgrade

• 测试所有环境的可视化，使用 fixtime 收集，查看摄像头的视角
• 完成专家策略的设计，特殊情况使用规则，常规情况使用 RL
• 完成所有场景的流量和特殊事件的设计
• 常规流量使用 RL 训练，获得 12*5 个强化学习模型

关于 route 车辆类型

    <vType id="background" length="5.00" color="220,220,220" tau="1.0"/>
    <vType id="police" length="5.00" color="blue" tau="1.0"/>
    <vType id="emergency" length="6.50" color="255,165,0" tau="1.0"/>
    <vType id="fire_engine" length="7.1" color="255,165,0" tau="1.0"/>
    <vType id="barrier_A" length="1" color="0,0,0"/>
    <vType id="barrier_B" length="1" color="0,0,0"/>
    <vType id="barrier_C" length="1" color="0,0,0"/>
    <vType id="barrier_D" length="1" color="0,0,0"/>
    <vType id="barrier_E" length="1" color="0,0,0"/>
    <vType id="tree_branch_1lane" length="1" color="0,0,0"/>
    <vType id="tree_branch_3lanes" length="1" color="0,0,0"/>
    <vType id="pedestrian" length="1" color="0,0,0"/>
    <vType id="crash_vehicle_1lane" length="1" color="0,0,0"/>
    <vType id="crash_vehicle_3lanes" length="1" color="0,0,0"/>
    <vType id="other_accidents" length="1" color="0,0,0"/>

场景收集

• France Massy
  • easy_random_perturbation
    • none
    • barrier
    • tree branch
    • pedestrian
    • crashed vehicle
  • easy_increasing_demand
    • none
    • barrier
    • tree branch
    • pedestrian
    • crashed vehicle
  • easy_fluctuating_commuter
    • none
    • barrier
    • tree branch
    • pedestrian
    • crashed vehicle

关于特殊场景说明

• 存在特殊车辆，需要优先级
• 某些情况导致无法通行
  • 行人/非机动车在路口跌倒或发生意外： 有人在路口中央摔倒或发生轻微碰撞，需要短暂处理，阻碍了车辆通行。
  • 路面突发障碍物： 如货物掉落（卡车上掉下箱子、油桶）、树枝/广告牌被风吹落、动物（如流浪狗）闯入路口中央等，导致车辆无法安全通过。
  • 临时交通管制： 因特殊警卫任务、大型活动、领导车队经过等，放置了路障，暂停了所有方向通行（即使显示绿灯）。

渲染画面 & QA

对应的标签是：

[
  {
    "question": "What are the vehicles in this lane doing? Are they moving or stopped?",
    "answer": "Vehicles are flowing through the intersection with green light."
  },
  {
    "question": "Does the image contain any emergency vehicles such as police cars, ambulances, or fire trucks?",
    "answer": "No, there are no emergency vehicles in the image."
  },
  {
    "question": "Is there any traffic accident or obstruction visible in the image?",
    "answer": "No, there are no visible traffic accidents or obstructions."
  },
  {
    "question": "How many lanes are there in total (including incoming and outgoing)?",
    "answer": "There are a total of 4 lanes, including 2 incoming lanes and 2 outgoing lanes."
  },
  {
    "question": "How many vehicles are visible in different areas of the image?",
    "answer": "Camera detects 5 vehicles total. incoming road: 2 clear vehicles nearby, 1 somewhat visible vehicles further out, and approximately 2 faint vehicles in the distance. Vehicles exiting the intersection are not counted;. Note: Distant vehicles may be less accurate due to limited visibility."
  },
  {
    "question": "What is the vehicle distribution across incoming lanes, considering visibility?",
    "answer": "Total 5 vehicles across 2 incoming lanes. Lane 0: 1 clear+1 faint+~2 very faint vehicles; Lane 1: 1 clear vehicles. Note: (1) Vehicles exiting the intersection are not counted; (2) Distant vehicles may be less accurate."
  },
  {
    "question": "What is the vehicle distribution across outgoing lanes, considering visibility?",
    "answer": "No vehicles detected on outgoing lanes."
  }
]

[
  {
    "question": "What are the vehicles in this lane doing? Are they moving or stopped?",
    "answer": "Vehicles are fully stopped at the red light."
  },
  {
    "question": "Does the image contain any emergency vehicles such as police cars, ambulances, or fire trucks?",
    "answer": "No, there are no emergency vehicles in the image."
  },
  {
    "question": "Is there any traffic accident or obstruction visible in the image?",
    "answer": "Yes, there is vehicle collision blocking the lane."
  },
  {
    "question": "How many lanes are there in total (including incoming and outgoing)?",
    "answer": "There are a total of 4 lanes, including 2 incoming lanes and 2 outgoing lanes."
  },
  {
    "question": "How many vehicles are visible in different areas of the image?",
    "answer": "Camera detects 23 vehicles total. incoming road: 10 clear vehicles nearby, 8 somewhat visible vehicles further out, and approximately 5 faint vehicles in the distance. Vehicles exiting the intersection are not counted;. Note: Distant vehicles may be less accurate due to limited visibility."
  },
  {
    "question": "What is the vehicle distribution across incoming lanes, considering visibility?",
    "answer": "Total 23 vehicles across 2 incoming lanes. Lane 0: 5 clear+4 faint+~3 very faint vehicles; Lane 1: 5 clear+4 faint+~2 very faint vehicles. Note: (1) Vehicles exiting the intersection are not counted; (2) Distant vehicles may be less accurate."
  },
  {
    "question": "What is the vehicle distribution across outgoing lanes, considering visibility?",
    "answer": "No vehicles detected on outgoing lanes."
  }
]

[jkmhhh]

4080S 渲染时间记录

结论：共享车辆模型后，low poly 和 high poly 车辆模型渲染时间不会相差很多。

修改 load_vehicles.py 前：

此时为独立加载车辆模型

high_poly 模型用时

low_poly模型用时：

修改 load_vehicles.py 后：

此时为共享车辆模型，不会单独加载

high_poly模型用时：

low_poly模型用时：

[wmn7]

特殊场景：

VLM 推理：

[wmn7]

[wmn7]

加入建筑物渲染 & 不同视角

渲染时间加倍，不需要每次都加建筑物渲染。

[wmn7]

https://bhpan.buaa.edu.cn/link/AAAE72441CD5574F638C8A8A900E14C29C
Folder Name: blenders
Expires: 2025-10-03 17:17