On the efficacy of texture analysis for crowd monitoring

Abstract

The goal of this work is to assess the efficacy of texture measures for estimating levels of crowd densities ill images. This estimation is crucial for the problem of crowd monitoring. and control. The assessment is carried out oil a set of nearly 300 real images captured from Liverpool Street Train Station. London, UK using texture measures extracted from the images through the following four different methods: gray level dependence matrices, straight lille segments. Fourier analysis. and fractal dimensions. The estimations of dowel densities are given in terms of the classification of the input images ill five classes of densities (very low, low. moderate. high and very high). Three types of classifiers are used: neural (implemented according to the Kohonen model). Bayesian. and an approach based on fitting functions. The results obtained by these three classifiers. using the four texture measures. allowed the conclusion that, for the problem of crowd density estimation. texture analysis is very effective.