Radar-tracking of low-observable targets such as drones suffers from low detection performance. In these type of applications, it is desirable to avoid data thresholding in order to preserve the weak target signal in the raw sensor data. This thesis considers the Multiple Object Tracking (MOT) problem in the context of radar Track-before-Detect (TrBD) processing, where the raw radar data is fed into the filtering process without previous compression into a finite set of detection/plots....

The Multiple Object Tracking problem for a known and constant number of closely-spaced objects in a track-before-detect context is considered. The underlying problem of decomposing a dataassociation- free Bayes posterior density is analyzed. A previously proposed solution for two objects moving in one-dimensional space is generalized for higher dimensional problems where t objects move in a M-dimensional space. The underlying problem is solved with the proposed Cross Modeling Tracker by means of hypothesizing physical crosses between the objects for a general t-MD objects case. In particular, the mathematical definition of cross-between-objects is generalized from a meaningful interpretation of the problem in the low dimensional setting. A method to provide optimal references for evaluation of the Cross Modeling Tracker is also considered. The Cross Modeling Tracker algorithm is validated with the optimal references by simulating t-MD closely-spaced objects scenarios. Wider applicability of the Cross Modeling Tracker with respect to comparable reviewed solutions is demonstrated via simulation experiments.