An effective modular approach for crowd counting in an image using convolutional neural networks.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Deep Learning

Background:

• Crowd counting accuracy is challenged by scale variation in images.
• Existing methods using dilated convolutions struggle with feature extraction and contextual information.
• Multi-scale feature extraction is often overlooked in cost-effective models.

Purpose of the Study:

• To propose a novel hierarchical dense dilated deep pyramid feature extraction (HDPF) method for single image crowd counting.
• To address limitations of standard dilated convolution (SDC) in capturing contextual information and multi-scale features.
• To enhance crowd counting accuracy by improving feature extraction and propagation.

Main Methods:

• Developed a HDPF method comprising General Feature Extraction Module (GFEM), Deep Pyramid Feature Extraction Module (PFEM), and Fusion Module (FM).
• Utilized densely connected dense stacked dilated convolutional modules (DSDCs) within PFEM for dense pixel sampling and contextual information.
• Employed a hierarchical structure for effective feature propagation across dilated convolutional layers (DCLs).

Main Results:

• The proposed HDPF method demonstrated effectiveness in extracting multi-scale information with an expanded receptive field.
• Dense connections in DSDCs enabled better contextual information acquisition compared to SDC.
• Simulations on Shanghaitech (Part-A, Part-B) and Venice datasets showed improved estimation accuracy.

Conclusions:

• The HDPF method offers a robust solution for single image crowd counting by effectively handling scale variation.
• The dense pyramid feature extraction approach significantly enhances the ability to capture relevant contextual information.
• The proposed technique shows promising results for improving crowd counting performance on benchmark datasets.