Dominant motion analysis in regular and irregular crowd scenes

In this paper we present a novel method for dominant motion analysis in crowded scenes, based on corner features. In our method, we initialize corner features on the scene, and advect them through optical flow. Approximating the moving corner features to individuals, their interaction forces, represented as endothermic reactions in a thermodynamic system, are computed using the enthalpy measure, thus obtaining the potential corner features of interest. These features are exploited to extract the orientation patterns, used as input priors for training a random forest. The experimental evaluation is conducted on a set of benchmark video sequences, commonly used for crowd motion analysis, and the obtained results are compared against other state of the art techniques.