IR/RF SPARK - IR/RF fusion using Stochastic Programming And Robust Kinematic features

SSCI and MIT team will approach a problem of fusing target data from sensor of different phenomenology using Probabilistic programming technology, Stochastic inference techniques based on Markov chain simulation, and Robust kinematic features. These methods will allow us to estimate the extend of information on metric, material, and kinematic properties of the observed low resolution targets available through infra-red and radar sensors. Any collaborating information available through both sensors would help their association and tracking between sensors of different phenomenology. Stochastic inference techniques based on Markov chain simulation will be employed to rapidly identify accurate interpretations of the data. Probabilistic programs go beyond classical pattern recognition techniques, statistical learning methods and Bayesian networks. These programs simulate hypothetical worlds according to the prior assumptions. Each possible execution path of the probabilistic program constitutes a distinct hypothesis over which Bayesian inference can be performed. In this framework, it becomes natural to capture detailed physical prior knowledge, even if the distributions involved are highly non-Gaussian and their time evolution is highly non-linear, making analytical representations intractable. The programs will also help the team assess the best wavebands to provide the richest set of potentially useful features for target characterization.