Track Correlation / Sensor Netting

The project objective is to develop a tracking algorithm that will fuse detections and tracks provided by a network of radar sensors with detections and tracks provided by a network of electro-optical (EO) sensors into multi-system tracks. The intent is to thereby achieve increased state estimation and classification accuracy in tracking ballistic missiles. Both target kinematic states and feature states are estimated. The target position-velocity vector (x,y,z,xdot,ydot,zdot) is modeled as 6-variate Gaussian. The state model accounts for Earth gravitation and atmospheric drag. The Extended Kalman Filter (EKF) is used for kinematic state estimation as an effective tool in dealing with nonlinearities in both the state equation and measurement equation. Target feature states such as radar cross section and color temperature are estimated using conjugate prior methods. Data association is treated as a classical assignment problem. The cost of a pairwise association is defined as the negative log-likelihood of that association. Repeated use of the Munkres algorithm is made to find the least-cost joint association, as well as pairwise associations that must be included in any joint association and pairwise associations that must be excluded from any joint association. Clustering methods are applied to detections and tracks whose associations are uncertain.