A Primer on Various Approaches to Data Association

To thread measurements (well, many call them “hits” or “plots”) of radar, sonar or imaging observations to a credible, smooth and reportable trajectory requires a filter. We’ll discuss those – Kalman, Unscented, particle, etc. – briefly. But the main topic here arises because one cannot even begin to filter without knowing which hits come from which targets, and which hits are complete nonsense (clutter). When wrapped inside some scheme for such data-association, a filter becomes a tracker. This talk is intended to explain, at a fairly high level, the intuition behind some of the popular tracking algorithms.