Skip to content
scdrand23Public

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

1 watching

Forks

Latest commit

 

History

21 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CVIPS: Connected Vision for Increased Pedestrian Safety

Implementation of the CVIPS project.

  • Dataset
  • CARLA data generation code

Dataset

The CVIPS dataset is available here.

Installation

Please check installation on how to setup CARLA for windows

Dataset Generation

Our dataset is generated using the CARLA simulator, providing diverse scenarios for collaborative perception.

Running the Data Generation Script

  1. Clone the repository:

    git clone https://github.com/cvips/CVIPS.git
    cd CVIPS
    
  2. Install the required dependencies:

    pip install -r requirements.txt
    
  3. Ensure CARLA is properly installed and running. Start the CARLA server in a separate terminal:

    /path/to/carla/CarlaUE4.exe
    
  4. Run the data generation script:

    python cvips_generation.py
    

    This script will connect to the CARLA server and generate the dataset based on the specified parameters.

Note: Make sure to adjust the CARLA server path and any configuration parameters in cvips_generation.py as needed for your specific setup.

Sample Examples

Here are some example commands to generate different scenarios:

  1. Four-way intersection in Town10HD with clear weather at noon:

    python cvips_generation.py --town Town10HD --intersection 4way --setup wsee --weather clear --time_of_the_day noon --crowd_level medium --crossing_percentage 0.1 --seed_value 23 --scenario scenario00001
    
  2. Four-way intersection in Town04 with rainy weather at sunset:

    python cvips_generation.py --town Town04 --intersection 4way --setup wwss --weather rainy --time_of_the_day sunset --crowd_level medium --crossing_percentage 0.1 --seed_value 42 --scenario scenario00002
    
  3. Three-way intersection in Town05 with clear weather at night:

    python cvips_generation.py --town Town05 --intersection 3way --setup wwss --weather clear --time_of_the_day night --crowd_level medium --crossing_percentage 0.1 --seed_value 73 --scenario scenario00005
    
  4. Four-way intersection in Town10HD with cloudy weather at noon:

    python cvips_generation.py --town Town10HD --intersection 4way --setup wwws --weather cloudy --time_of_the_day noon --crowd_level medium --crossing_percentage 0.1 --seed_value 101 --scenario scenario00011
    

These examples demonstrate various combinations of towns, intersection types, weather conditions, times of day, and camera setups. Users can modify these parameters to generate different scenarios as needed.

Sample Visualizations

We provide visualizations to demonstrate different perspectives in our dataset: (this can be done using Code/data_viz.py)

  1. Infrastructure View (6 cameras) Infrastructure View This visualization shows the scene from fixed infrastructure cameras, providing a comprehensive view of the intersection from multiple angles.

  2. Vehicle View (6 cameras) Vehicle View This visualization demonstrates the perspective from a moving vehicle, showcasing the view from various angles around the vehicle.

  3. Collaborative View (4 cameras) Collaborative View This visualization combines views from both vehicles and infrastructure, illustrating the collaborative perception aspect of our dataset.

  • Multiple camera perspectives (infrastructure, vehicle-mounted, and collaborative)
  • Various weather conditions and times of day ()
  • Diverse traffic scenarios with multiple vehicles and pedestrians setup (controlled by seeds, number of pedestrians, number of vehicles to be spawned)

For Detection in collaborative setup check the documentation of V2XFormer

First, prepare the dataset with this format data_preparation . Then, use V2XFormer according to the instructions here

  1. Installation
  2. Training and Evaluation -- adjust the datapaths and number of classes accordingly

Acknowledgement

This project is based on the following open-source projects: BEVerse, Fiery, open-mmlab, and DeepAccident.

Citation

If you find this dataset or code useful, please cite:

@techreport{cvips2024,
  title     = {Connected Vision for Increased Pedestrian Safety (CVIPS)},
  author    = {Bhagavatula, Vijayakumar and Sankaranarayanan, Aswin and Shenkut, Dereje},
  institution = {Carnegie Mellon University, Safety21 University Transportation Center},
  year      = {2024},
  month     = {July},
  url       = {https://rosap.ntl.bts.gov/view/dot/77491}
}

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

1 watching

Forks

Releases

Packages

Used by

Contributors

Languages