A Novel and Integrated Framework for Sensor Registration and Sensor Management

In the registration part, the goal is to eliminate sensor biases and yield an accurate and integrated picture of the target tracks. In the sensor management part, the goal is to use a limited number of sensors to achieve certain tasks (e.g., accurate fire control) without using computationally expensive methods. We propose a highly innovative approach to sensor registration and sensor management. Our architecture can deal with disparate sensors. In the sensor registration part, the first step is to initiate the tracks using sensor measurements. We propose a recursive nonlinear prediction approach to perform this initialization. This predictor is robust to background clutter. Then we propose a track correlation technique called sequential minimum normalized distance nearest neighbor (SMNDNN) method. We have shown that only those track pairs which satisfy the minimum normalized distance will remain, and the maximum possible assignment is arrived at under the condition that each track is correlated only once. Finally, we apply Expectation-Maximization-Extended Kalman Filter (EM-EKF) to yield combined target tracks. In the sensor management part, we propose to apply recently developed sensor management algorithms known as covariance control, which has been proven to be very useful in estimating interacting (close) target tracks using a small number of sensors. The computational burden is also reasonable.