/social-distance-tool-with-depth

Basic implementation of a social distance violation detection on any generic video using faster_rcnn from Facebook's Detectron2 and incorporating depth estimation from monodepth2 for more accurate results.

Primary LanguageJupyter Notebook

Social Distance Tool with depth

example input output gif

This tool combines two algorithms to accurately detect people who are violating the social distancing protocol:

  • Facebook/Detectron2 (Faster RCNN implementation)https://github.com/facebookresearch/detectron2
  • "Digging into Self-Supervised Monocular Depth Prediction" https://github.com/nianticlabs/monodepth2

Starter code taken from an excellent tutorial from Aravind Pai: https://www.analyticsvidhya.com/blog/2020/05/social-distancing-detection-tool-deep-learning/

Use:

Social-Distance-Tool-with-Depth.ipynb

Libraries needed:

  • Detectron2 = 0.13
  • OpenCV >= 3
  • Matplotlib
  • tqdm
  • pytorch = 1.4
  • torchvision = 0.4

Input:

  • A video sequence

Output:

  • bounding boxes on all persons detected in the video
  • highlighing people who are in close proximity
  • depth map for accurate calculations

This code is for non-commercial use.

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.