Estimation of crowd density based on wavelet and support vector machine

Automatic estimation of crowd density is very important for the safety management of the crowds. In particular, when the density of the crowds exceeds a critical level, the safety of people in the crowd may be compromised. This paper describes a novel method to estimate the crowd density based on the combination of multi-scale analysis and a support vector machine. The algorithm will first transform the crowd image into multi-scale formats using wavelet transform. The first-order and second-order statistical features at each scale of the transformed images are then extracted as density character vectors. Furthermore, a classifier based on a support vector machine is designed to classify the extracted density character vectors into different density levels. Compared with the conventional statistical techniques and wavelet energy techniques used in single-scale images, the test results of a set of 300 images show that the proposed algorithm can achieve much improved performance and more detailed information of the crowd density can be captured by the new feature extraction method.