Expectation-Maximization Algorithm for Radar Centroid Processing with Tracker Feedback (EMARCT)

The raw detection data obtained by a radar typically consists of primitive measurements that correspond to discrete range-bearing cells, multiple of which can contain energy reflected from a single target. The radar system employs a centroid processing algorithm to combine the primitive measurements and form a single detection report. However, if multiple targets exist close to one another, this centroid process can produce undesirable results such as track degradation and/or track drop, since the measurements for more than one target may be combined to produce a single report. We propose a method for utilizing feedback from the tracker to produce the appropriate number of centroid estimates in a given area, thus avoiding track drop. Our method is a modification of the Expectation-Maximization algorithm as applied to the Gaussian Mixture Estimation problem; accordingly, we will refer to our algorithm as EMARCT (an Expectation Maximization Algorithm for Radar Centroid processing with Tracker feedback). In Phase II, we will enhance the algorithm by improving the initialization and termination components of the iteration and by adding the ability to independently estimate the number of targets. We will program the algorithm in C++ and demonstrate its feasibility on simulated and real-world data.