AbstractOn Some New Methodologies for Joint Detection and Trajectory Estimation of Moving Objects In this talk, a dynamical system denotes any object evolving in a (discrete) state space: examples include - but are not limited to - manoeuvering targets describing trajectories with time-varying ranges andor azimuths, andor elevations , as well as users logging in and out a communication network - possibly wireless - and experiencing time-varying channel states. Under these circumstances, detection and estimation are inherently coupled, since detecting the presence of one or more objects inevitably requires estimating their trajectories: all in all, the problem boils down to trajectory estimation under uncertainty as to the object(s) presence and number. Some additional data that can be available are side information as to the admissible target trajectories, encapsulated, in a Bayesian setting, in state-transition probability laws and, in a non-Bayesian one, by limitations to the possible values of the motion parameters (velocity, acceleration, and so on). The scenario may be a truly challenging one as the number of objects may be itself time-varying, possibly randomly: indeed, the intuitive procedure of splitting up the problem in two distinct phases, wherein detection comes first and estimators operate only on declared objects, is sub-optimal and may lead to heavy losses as compared to a unified “optimal” procedure operating in a single step. This talks is aimed at illustrating some recent methodologies that have been proposed , and sometimes implemented, with an eye kept to the central issue of maintaining the computational complexity at a level compatible with real-time implementation. |