Design and Implementation of Intelligent Crowd Monitoring System

Abstract

One of the key objectives of crowd management is to plan and design a suitable strategy to equalize the crowd flow, prioritizing safety of the public. People tend to behave differently in a crowded scenario. For instance, if there is an altercation between angry fans or a fire in a building and in several other unforeseen circumstances panic and riot- like behaviour can break out. Therefore, an effective crowd management system is required to provide real time analysis of the crowd to law enforcement agencies for faster and efficient decision making.  In orderto achieve this, a network for Congested Scene Recognition known as CSRNet is introduced to deliver a data oriented and deep learning method that interprets densely-packed areas in order to obtain a precise crowd-count estimate. It generates high resolution and accurate density maps for a given input image. In this paper, CSRNet is implemented on the Shanghai-Tech data set and the MAE (Mean Absolute Error) obtained is significantly about 25% lower than some of the traditional approaches usedpreviously.