A new approach of crowd density estimation

Abstract

Crowd density estimation is important in crowd analysis, this paper proposes a new approach used for crowd density estimation. First, background is removed by using a combination of optical flow and background subtract methods. Then according to texture analysis, a set of new feature is extracted from foreground image. Finally, a self-organizing map neural network is used for classifying different crowds. Some experimental results show compared to former crowd estimation methods, the proposed approach can carry out the estimation more accurately, the rate of true classification is 85.6% on a data set of 500 images.