Year: 2026
Type: article
Abstract: Crowd counting, in accurate terms, is crucial for applications such as video monitoring and traffic flow control. While convolutional neural networks have significantly improved the precision of deep learning, challenges remain due to variability in crowd size and complex backgrounds. This research addresses these issues by introducing the Multi-Scale Residual Network (MSRR), a deep learning framework that incorporates a dimension attention mechanism. MSRR captures long-range relationships across feature dimensions, further improving its ability to handle complex scenes. Tested on the benchmark ShanghaiTech crowd counting datasets, MSRR outperforms existing state-of-the-art methods, achieving at least a 5.7% improvement in mean squared error and mean absolute error. This marks a major advancement in crowd counting techniques, offering an effective solution for complicated scenarios.
Institution: Rama University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid:
OpenAlex ID: W7118190229