An Expectation Maximization Based Simultaneous Registration and Fusion Algorithm for Radar Networks

In this paper, we present an expectation maximization (EM) based simultaneous registration and fusion algorithm for multiple radars network. This simultaneous registration and fusion approach has advantages over other track registration techniques such as augmented Kalman filtering approach. Systematic biases (including radar time bias, radar two angular biases, and radar range bias) are estimated using the EM algorithm. The EM algorithm can guarantee that the estimated systematic biases can converge to a stationary point of the maximum likelihood function. In order to track maneuvering targets, three kinematic models are used to have a more complete description of the motion of a target: the constant velocity (CV), the constant acceleration (CA), and the coordinated turn (CT) models. The conditional expectations and covariances of the system states can be effectively computed by interactive multiple model (IMM) approach. The IMM method is a recursive hybrid filtering technique that provides a good balance between performance and complexity. We employ a fixed-interval smoother-based IMM approach to estimate the system states. The forward and backward Kalman filters are used to obtain a smooth estimate. The IMM approach is combined with the EM algorithm for simultaneous registration and fusion. The proposed algorithm consists of two steps: the first step is to adopt the fixed-interval smoother-based IMM method to calculate the conditional expectation of the log likelihood function using the current estimate of the systematic biases and the observations; the second step updates the parameter estimate by maximizing the log likelihood function