State estimation using GM-PHD filter applied to the tracking of individuals

The target tracking problem becomes more realistic when factors as time varying target number, measurements immersed in clutter and/or false alarms are introduced in the problem modeling. The Random Finite Sets theory provides an appropriate framework to deal with multiple targets in a dynamic scenario for the mathematical modeling and stochastic filtering. This paper analyzes the tracking problem of walking/running people in a surveillance system using the Gaussian Mixture PHD Filter. This filter is specially suited to cope with multiple targets in presence of dense cluttering. In order to observe the filter behavior in front of a large number of individuals, a random path generator based on simple dynamics is devised, comprising birth and spawned trajectories.