SCIENCE@home

DACNet: A Density-Adaptive Counting Network for Real-World Crowd Analysis Without Overhead

Jianping Yue,Bohuan Xue,Wenli Wu,Rui Fan,Xiaoyu Tang

Published 2026 in IEEE Sensors Journal

ABSTRACT

In practical applications of crowd counting, the density and scale of human heads often vary significantly due to the influence of the camera’s perspective effect. Pointbased methods fail to consider crowd density variations and face issues with inaccurate matching. Inspired by the spatial perception function of the posterior parietal cortex in the human brain, this article proposes a density-adaptive counting network (DACNet), which assists object counting through auxiliary points. First, we propose a lightweight detail enhancement Mamba block (DEmamba Block), which combines convolution and state space models (SSMs) to enhance blurred details in densely crowded regions. Second, we propose a plug-and-play adaptive channel focus module (ACFM). ACFM introduces a channel weight selection algorithm, leveraging the advantages of multiple weights. Finally, we propose a density-adaptive auxiliary point guidance (DA-APG) strategy in the detection head. DA-APG generates positive and negative auxiliary points at varying distances around the ground truth points based on crowd density as additional supervisory signals, addressing the issue of crowd density variation. Moreover, this DA-APG strategy is only applied during training, and does not incur additional computational cost. To facilitate research on crowd density variations in real-world scenarios, we introduce a specialized dataset named VariDensity-CC. Experiments on nine datasets show that DACNet achieves the best overall balance between accuracy and speed. Furthermore, DACNet has been deployed on edge computing devices for real-world testing and demonstrates real-time performance. The code and dataset are available at: https://github.com/SCNURISLAB/DACNet

PUBLICATION RECORD

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-64 of 64 references · Page 1 of 1

CITED BY

  • No citing papers are available for this paper.

Showing 0-0 of 0 citing papers · Page 1 of 1