A deep crowd density classification model for Hajj pilgrimage using fully convolutional neural network

Area of Science:

• Computer Vision
• Artificial Intelligence
• Data Science

Background:

• Traditional crowd analysis methods struggle with accuracy for distant crowds in surveillance footage.
• High-density crowd analysis is crucial for managing large events like the Hajj and Umrah pilgrimages.

Purpose of the Study:

• To enhance crowd analysis and density prediction for Hajj and Umrah pilgrimages.
• To overcome limitations in estimating crowd density from distant surveillance cameras.

Main Methods:

• A fully convolutional neural network (FCNN)-based approach was developed for crowd density classification.
• A new dataset, Hajj-Crowd-2021, was created based on Hajj pilgrimage scenarios.
• The proposed model was validated against existing models and datasets (UCSD, JHU-CROWD).

Main Results:

• The FCNN-based method achieved 100% accuracy on the Hajj-Crowd-2021 dataset.
• High accuracies of 98% and 98.16% were obtained on the UCSD and JHU-CROWD datasets, respectively.
• The proposed model and dataset outperformed state-of-the-art methods in most evaluations.

Conclusions:

• The FCNN-based approach significantly improves crowd density estimation, especially in challenging, high-density scenarios.
• The Hajj-Crowd-2021 dataset provides a valuable resource for advancing crowd analysis research.
• This research offers a robust solution for crowd monitoring in large-scale religious gatherings.