Enabling netted sensor fusion for anti-submarine warfare in uncertain and variable environments - MP 60-10

This proposal describes an approach for multi-sensor, multi-target SR & G based on theoretically sound Bayesian probability. A key aspect of our approach is employment of a Rao-Blackwell (marginalized) particle filter to"jointly"track all relevant state parameters and their uncertainty. The filter state parameters are sensor biases, navigation, biases, sensing platform kinematics (position and velocity), as well as the state (position, velocity and classification) of tracked targets. Sophisticated filtering algorithms are required since the bias and measurement uncertainties impact the system dynamics in complicated, nonlinear ways: hence, traditional Kalman filters are inappropriate. A particle filter represents an arbitrary state probability distribution and does not require a linear system model. Particle filtering methods are ideal for systems with highly nonlinear dynamics and high levels of uncertainty. Furthermore, particle filters are readily extensible to discrete and categorical parameters such as class labels. A second key innovation is the use of target feature information and classification information as well as kinematic information to perform the track-to-system data association. The MAP association over all particles will feed forward/backward to the joint particle filter to perform joint target/bias tracking.