Improving algorithms for artificial intelligence-based crowd density estimation and public safety assurance

Abstract

The article examines the automatic estimation of crowd density from video imagery in places of mass gathering, as well as the task of converting the resulting estimate into a management decision aimed at ensuring public safety. The stages in the development of density estimation algorithms are analysed comparatively, and the limitations of modern deep-learning-based solutions in terms of accuracy, computational resources and adaptability to other environments are demonstrated on the basis of sources published in 2025-2026. A four-stage algorithmic scheme is proposed, comprising modules for camera calibration, density map generation, risk level classification and the issuing of recommendations on force deployment, and its operation is demonstrated by means of a numerical example. The conditions for implementing the system within the current legal and regulatory requirements of the Republic of Uzbekistan in the field of artificial intelligence are also substantiated.