Dynamic Adversarial Risk Estimation Based on Labeled Multi-Bernoulli Tracker

Abstract

In many military and civilian areas, there exists a scenario in which multiple intruders from an adversary attempt to enter important region of our own to carry out intentional malign activity. Adversarial Risk (AR) estimation is used to assess and predict the expected damage to our valuable assets from the actions of online adversaries based on measurements performed by sensors. To evaluate random and time-varying AR, this study proposes a stochastic AR estimation approach based on a Labeled Multi-Bernoulli (LMB) tracker. First, in the formulation of LMB filtering, expressions of the minimum mean squared error estimation of the stochastic AR are derived for the additive and multiplying model. Second, by combining the Gaussian mixture and sampling approximations, we devise a numerical calculation approach for the proposed AR estimations. Third, we achieve an online evaluation of the expected damage to our valuable assets from the adversary by embedding the proposed AR estimation and LMB filtering. The effectiveness and performance advantage of the proposed estimation algorithms are verified using measurements from radars, considering a simulated scenario wherein multiple lethal targets hit the radar positions.

Keywords

• adversarial risk estimation
• multiple-target tracking
• labeled multi-bernoulli filter
• random finite set statistics
• threat level assessment
• posture reconstruction