Visual Crowd Detector

Application detecting gatherings of people on images. It's goal is to help fight against COVID-19 by pinpointing the most crowded spots. It was created while participating in https://www.hackcrisis.com/ hackathon.

The application supports operators of monitoring systems to help them control if people don't gather in larger groups on what areas should be disinfected in particular.

Example of usage

People profiles marked in red marks gatherings. Red dots represents points where a lot of people have been passing. At the bottom total number of people in the image is shown.

A demo video is available on YouTube: https://www.youtube.com/watch?v=vq6eWuhRTwI

Environment preparation

We strongly recommend docker usage, as it helps to keep dependencies for different project separate. If you prefer not to use it, install the dependencies from by running: pip install -r requirements.txt.

Whole development environment is managed by Makefile. There are the following commands:

• make build - creates docker image and installs required dependencies
• make dev - starts the container and give access to the console
• make dev_gui - start the container allowing for running window applications
• make lab - starts the image and jupyter lab environment

Before use

You need to download the weights for the network from: http://download.tensorflow.org/models/object_detection/mask_rcnn_inception_v2_coco_2018_01_28.tar.gz

Usage

For a fast start with default values use: python detector.py. More details about parameters are provided below:

NAME
    detector.py - Main loop of the application. Iterates through all frames in the input video detects people and marks dangerous regions.

SYNOPSIS
    detector.py <flags>

DESCRIPTION
    Main loop of the application. Iterates through all frames in the input video detects people and marks dangerous regions.

FLAGS
    --model_path=MODEL_PATH
        Path to the inference graph.
    --video_path=VIDEO_PATH
        Path to the input video