Multiarget tracking (MTT here after) deals with state estimation of an unknown number of (moving) targets. Available measurements may have originated from the targets if they are detected or of a special model called "clutter". Clutter is generally considered as a model describing false alarms. Its (spatio-temporal) statistical properties are quite different from target ones which render possible the separation of target tracks on the one hand and clutter model on the second. To perform multitarget tracking the observer has at its disposal a huge amount of data, possibly collected on multiple sensors. Elementary measurements are receiver outputs; e.g. bearings, ranges, time-delays, dopplers, etc. But the main difficulty comes from the assignment of a given measurement to a target model. For critical situations, these assignments are unknown, as are the true target models. This is a neat departure from classical estimation problems. Two distinct problems have to be solved jointly: data association and estimation. Since the mid-sixties, this subject has attracted considerable attention. This interest is due to the challenging difficulty of the problem as well as to the operational requirements and computation possibilities. The simpler approach is probably the Nearest Neighbor approach. However, despite its simplicity, it has fundamental limitations. A natural framework is to face the combinatorial complexity of the data association problem. Pioneering work in this way has been made by Morefield, using integer programming. More recently, this approach has been greatly improved by means of approximate methods (Lagrangian duality) for solving the combinatorial assignment problem. Other approaches are more relevant to the general class called Multiple Hypotheses Tracker (MHT), in wlu'ch measurements received at a scan are assigned to intialized tracks, new targets or false alarms. Pruning and gating techniques are used to retain the most likely hypotheses and limit their number. By this way, combinatorial explosion may be strictly contained inside reasonable bounds. However, there is a strong risk to eliminate correct assignments.
展开▼
