Deep-Learning-Based Crowd Density Estimation: A Comprehensive Survey

Abstract

The importance of crowd density estimate (CDE) in event management, urban planning, and public safety has made it a crucial area of study in computer vision. Conventional computer vision techniques are susceptible to challenging situations in the crowd including occlusions, scaling, and light conditions. Since the implementation of deep learning, convolutional neural networks (CNNs) and, more recently, transformer-based and hybrid designs, have demonstrated state-of-the-art performance in the estimate of crowd density and person count. The paper will provide an extensive review of the deep-learning-based methods of crowd density estimation published between 2015 and 2025 in terms of architectural development, the variability of the datasets, and the advancement of the methodology. The trends, benchmarks, and challenges are consolidated in the survey, which will serve as a roadmap in the future research of deep-learning-based crowd analysis.