A novel method of crowd estimation in public locations

Abstract

A novel method of crowd estimation is proposed in this paper: Firstly, surveillance image is divided into bit planes by OSTU algorithm, the pixel ratio of foreground to background and complexity of bit planes are taken as feature vectors of crowd estimation. The degree of crowd density of the scene is classified into several grades, BP neural network is used for training and then the classification model is constructed, through which the estimation of crowd density can be obtained. Experiments were taken based on video of two real scenes, the result show that this proposed approach is able to judge the levels of congestion with accuracy higher than 85%.