Application of invariant moments for crowd analysis

Download

PDF Download (4Mb)

Description/Abstract

The advancement in technology such as the use of CCTV has improved the effects of monitoring crowds. However, the drawback of using CCTV is that the observer might miss some information because monitoring crowds through CCTV system is very laborious and cannot be performed for all the cameras simultaneously. Hence, integrating the image processing techniques into the CCTV surveillance system could give numerous key advantages, and is in fact the only way to deploy effective and affordable intelligent video security systems. Meanwhile, in monitoring crowds, this approach may provide an automated crowd analysis which may also help to improve the prevention of incidents and accelerate action triggering. One of the image processing techniques which might be appropriate is moment invariants. The moments for an individual object have been used widely and successfully in lots of application such as pattern recognition, object identification or image reconstruction. However, until now, moments have not been widely used for a group of objects, such as crowds. A new method Translation Invariant Orthonormal Chebyshev Moments has been proposed. It has been used to estimate crowd density, and compared with two other methods, the Grey Level Dependency Matrix and Minkowski Fractal Dimension. The extracted features are classified into a range of density by using a Self Organizing Map. A comparison of the classification results is done to determine which method gives the best performance for measuring crowd density by vision. The Grey Level Dependency Matrix gives slightly better performance than the Translation Invariant Orthonormal Chebyshev Moments. However, the latter requires less computational resources.