Social-Distancing

Social-Distancing is an open-source project for automatically estimating interpersonal distance from uncalibrated RGB cameras. The software can be freely used for any non-commercial applications to assess compliance with safe distances. The code is open and can be improved with your support, please contact us at [email protected] if you would like to help us.

What's New

[December 18th, 2020]

• The dataset can now be downloaded from the following link: social_distancing_dataset.

[November 5th, 2020]

• Our work have been accepted at WACV 2021. Chek out the paper!

[November 4th, 2020]

• Alghorithm updates with better distance evaluation and computational speed up
• Fast ellipses intersections check with Shapely
• Added masking support to select interesting areas
• Streaming support
• Acquisition from Jetson nano camera
• Ubuntu 20.04 with Cuda 10.1 support

[April 24th, 2020]

• Code for live camera acquisition and video processing.
• New video samples in the samples folder.

Description

Given a frame captured from a scene, the algorithm first detects visible people in the scene using an off-the-shelf body pose detector and estimates the height of the people through measuring the distance from their body joints. In the second step, the algorithm estimates an area of one meter around all the detected people. This distance is roughly estimated proportional to a typical human body height of 160 cm and can be used to draw a circle centered in human position in the scene. In the third step, the Homography of the scene is estimated given two parameters which essentially map the rectangular bird’s view model for the scene to the trapezoidal perspective view of the scene. These two parameters need to be manually tuned to estimate best the scene perspective. According to the Homography matrix, the safe circular distance for each person is converted to ellipsoids in perspective view. The people are considered to be staying in safe distance from each other if their ellipsoids do not collide. Conversely, if ellipsoids of two people collide, those people are considered as being in risk and their ellipsoids will be shown in red.

If you use this code as part of your research, please cite our work.

@inproceedings{vsd2021,
   title={Single Image Human Proxemics Estimation for Visual Social Distancing},
   author={Aghaei, Maya and Bustreo, Matteo and Wang, Yiming and  Bailo, Gian Luca and Morerio, Pietro and Del Bue, Alessio},
   booktitle={IEEE Winter Conference on Applications of Computer Vision (WACV)},
   year={2021}
}

Installation steps

Code is developed in Python3 and tested on Ubuntu 20.04 with NVidia driver, Cuda 10.1 and Cudnn 7.6.5.

• [x] Install the requirements
  To run this code, you need to install:

  • OpenPose 1.6.0:
    Please follow the instruction in the repository gitHub and install OpenPose in social-distancing/openpose/ folder.
    In case you prefer to use a different OpenPose installation folder, you can pass it using the --openpose_folder argument.

  • OpenCV:
    apt-get install python3-opencv
pip3 install opencv-python

  • PyTurboJPEG:
    pip3 install PyTurboJPEG

  • Shapely: pip3 install Shapely

  • Itertools: pip3 install itertools

  • Numpy: pip3 install numpy

Usage

python3 process_source.py -h  #help

Images

python3 process_source.py --image_in <path to the input image> --image_out <path to the result image to be saved> -- background_in <path to the background> --horizontal_ratio 0.7 --vertical_ratio 0.7

Videos

python3 process_source.py --video enabled --stream_in [path to the input video] --stream_out [path to the result video] --horizontal_ratio 0.7 --vertical_ratio 0.7

Network stream

python3 process_source.py --preview disabled --streaming enabled --video_port [port] --js_port [js_port] --stream_in [ address ]

Dataset

The dataset can now be downloaded from the following link: social_distancing_dataset (24.5 GB).

Disclaimer

Information provided by the software is to be intended as an indication of safe distance compliance. It is not intended to measure the actual metric distance among people.

The software is provided "as is", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose and noninfringement. In no event shall the authors, PAVIS or IIT be liable for any claim, damages or other liability, whether in an action of contract, tort or otherwise, arising from, out of or in connection with the software or the use or other dealings in the software.

LICENSE

This project is licensed under the terms of the MIT license.

This project incorporates material from the projects listed below (collectively, "Third Party Code"). This Third Party Code is licensed to you under their original license terms. We reserves all other rights not expressly granted, whether by implication, estoppel or otherwise.

1. OpenPose
2. OpenCV