Large-Scale Crowd Counting by Hierarchical Context Guided Aggregation Network

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Convolutional Neural Networks (CNNs) show strong feature learning but struggle with scale variation and complex backgrounds in crowd counting.
• Existing CNN-based methods face limitations in achieving optimal performance for large-scale crowd scenes.

Purpose of the Study:

• To propose the Hierarchical Context-Guided Aggregation Network (HCGANet) for accurate crowd density map generation.
• To address the challenges of scale variation and complex backgrounds in crowd counting.

Main Methods:

• HCGANet gradually aggregates crowd region information for precise density maps.
• A Multiscale Context Extraction Module (MCEM) extracts contextual information at each level, reducing feature fusion issues.
• A Hierarchical Context Aggregation Module (HCAM) fuses semantic and detailed information across levels.
• An Attention-Guided Module (AGM) focuses the network on crowd regions for enhanced accuracy.

Main Results:

• Comprehensive experiments on various datasets demonstrate HCGANet's effectiveness.
• HCGANet achieves more accurate crowd counting across different scenes compared to state-of-the-art methods.
• The proposed network successfully generates crowd density maps that closely approximate actual crowd distribution.

Conclusions:

• HCGANet offers a significant advancement in CNN-based crowd counting.
• The network's hierarchical and attention-guided approach effectively mitigates challenges in large-scale crowd scenes.
• HCGANet provides a robust solution for accurate crowd density estimation in diverse environments.