Social-Distance-Breach-Detector-OpenCV-DL

OpenCV with Python based project using Caffe Deep Learning Framework and Triangle Similarity Theorem to identify Social Distance Breaches.

Single Shot MultiBox Detection

Utilizes SSD Framework to perform Object Detection. SSD works on a base network in this case - MobileNet to extract feature maps and applies Convolution Filters to make predictions.

Model

Caffe Deep Learning Framework (Single Shot MultiBox Detection) to identify people in frame - SSD_MobileNet.caffemodel. For more information on Caffe

Distance and Depth

Triangle Similarity Theorem to measure distance from objects to camera. Further, centroids of each bounding box are taken as reference to find distance between two objects.

Streaming

The Flask App was created as a way to show portability of the project. Given the lightweight MobileNet, we could receive a direct video surveillance on live feed with a lag of approx 0.25 seconds.

Steps to Execute Flask-App Code

Launch terminal
cd to Project Directory
Write the following commands to run :
- On Windows CMD:
  
  set FLASK_APP=main.py
  
  flask run --host=0.0.0.0
- On Linux Terminal:
  
  export FLASK_APP=main.py
  
  flask run --host=0.0.0.0
- On Windows Powershell:
  
  $env:FLASK_APP=main.py
  
  flask run --host=0.0.0.0
Go on to preferred web browser : 'localhost: <port number being shown on terminal window>'
To exit from stream and terminate operations -> ctrl + c