VCNN4PuDe: A Novel Voronoi-based CNN Framework for Pushing Person Detection in Crowd Videos

---

This repository is for the submited paper:

Alia, Ahmed, et al. "A Novel Voronoi-based Convolutional Neural Network Framework
for Pushing Person Detection in Crowd Videos". 2023

Table of Contents

• Goal
• Motivation
• Architicture of VCNN4PuDe
• Codes of VCNN4PuDe
• Samples
• Framework Installing
• Framework Running
• Codes for CNN Architectures and Training
• Trained CNN models
• Test Sets
• Codes for Trained CNN Models Evaluation

Goal

The main goal of this article is to introduce a framework (**VCNN4PuDe**) for identify the persons who engage in pushing within videos of crowds.

Motivation

Detecting pushing persons within videos of crowded event entrances is crucial for understanding pushing dynamics, thereby designing and managing more comfortable and safer entrances.

Architicture of VCNN4PuDe Framework

Codes of VCNN4PuDe Framework

Codes of VCNN4PuDe

Samples

Input video with its trajectory data

You can access them by this link.

Note: They were taken from Pedestrian Dynamics Data Archive hosted by FZJ.

Annotated Video produced by VCNN4PuDe Framework

VCNN4PuDe Installing on Google Colab

1. Create a directory named VCNN4PuDe on your drive.

2. Access VCNN4PuDe directory.

3. Add new notebook and run the follwing commands

   a. Mount Google Drive

   from google.colab import drive
   drive.mount('/content/gdrive')
   

   b. Access VCNN4PuDe directory Folder

   %cd /content/drive/My Drive/VCNN4PuDe/
   

   c. Clone VCNN4PuDe Framework

   git clone https://github.com/abualia4/VCNN4PuDe.git
   

   d. Install keras-preprocessing module

   !pip install keras-preprocessing
   

VCNN4PuDe Running

Open the run notebook and follow the instructions in the notebook, and the annotated Video.mp4 will be stored in the annotated folder.

Note: If some libraries are required for running the framework, use the following command to install it

!pip install module/library name

Codes for CNN Architectures and Training

Open the notebook

Trained Models

All trained models produced in this article are available at this link

Test Sets

Two test sets are available at this link

Codes for Trained Models Evaluation

Two test sets are available at this link

Acknowledgement

1. Thanks to the authors of voronoi_finite_polygons_2d function.
2. Thanks to the author of Create_random_polygon class.

citation

Soon