Modelling of Objects Behaviour for Their Re-identification in Multi-camera Surveillance System Employing Particle Filters and Flow Graphs

Abstract

An extension of the re-identification method of modeling objects behavior in muti-camera surveillance systems, related to adding a particle filter to the decision-making algorithm is covered by the paper. A variety of tracking methods related to a single FOV (Field of Vision) are known, proven to be quite different for inter-camera tracking, especially in case of non-overlapping FOVs. The re-identification methods refer to the determination of the probability of a particular object's identity recognized by a pair of cameras. An evaluation of the proposed modification of the re-identification method is presented in the paper, which is concluded with an analysis of some comparison results brought by the methods implemented with and without a particle filter employment.