Moving object detection and respectively tracking are two intermediary steps for automatic scene understanding and generally result in complicated spatio-temporal patterns of high-level information. These patterns often require statistical analysis for which appropriate statistical tools and suitable mathematical models have to be developed. Stochastic geometry is one of the areas of mathematical research which seeks to provide such methods and models. Stochastic geometry approaches rely on a complete simulation of complex densities of interacting spatio-temporal processes, using Monte Carlo techniques.

In this talk, I will present a new spatio-temporal point process model specifically developed for multiple (moving) object tracking in wide area aerial surveillance videos. I will describe how this model can be efficiently simulated using the Reversible Jump Markov Chain Monte Carlo (RJMCMC) sampler and I will show results on different data sets acquired using sub-orbital and orbital surveillance systems.