Deep Dilated Convolutional Neural Network for Crowd Density Image Classification with Dataset Augmentation for Hajj...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Crowd Dynamics

Background:

• The annual Hajj pilgrimage in Mecca attracts millions of Muslims globally.
• Managing massive crowds poses significant safety and logistical challenges.
• Accurate crowd density estimation is crucial for preventing accidents and ensuring pilgrim safety.

Purpose of the Study:

• To introduce a deep learning model for accurate crowd density analysis in the Hajj pilgrimage context.
• To develop an effective data augmentation technique for Hajj-specific crowd scenarios.
• To provide a robust solution for crowd density measurement using distant surveillance cameras.

Main Methods:

• Developed a deep Hajj crowd dilated convolutional neural network (DHCDCNNet) for crowd density analysis.
• Implemented a data augmentation technique with two routes: magnitude/polar magnitude and morphological/skeleton transformation.
• Utilized a single framework for extracting both high-level and low-level features for comprehensive analysis.

Main Results:

• The DHCDCNNet achieved high accuracy across multiple datasets: 97% (JHU-CROWD), 89% (UCSD), and 100% (Hajj-Crowd).
• VGGNet and ResNet50 approaches also demonstrated strong performance on the Hajj-Crowd dataset (98% and 99% accuracy, respectively).
• The model effectively handles high-density scenarios (7-8 persons/sqm) common in the Tawaf area.

Conclusions:

• The proposed DHCDCNNet offers a highly accurate and reliable method for crowd density analysis during the Hajj.
• The augmentation technique successfully generated a valuable dataset for Hajj pilgrimage scenarios.
• This research contributes to enhanced crowd management strategies for large-scale religious gatherings.