Multi-sensor multi-target joint tracking and classification

To account for joint tracking and classification (JTC) of multiple targets from a sequence of noisy and cluttered observation sets under non-detection, this paper proposes a recursive JTC algorithm of model-class-matched probability hypothesis density (PHD) filter with the particle implementation, i.e., MCM-PHD-JTC. Assuming that each target class has a class-dependent kinematic model set, a model-class-matched PHD filter (MCM-PHD) is assigned to each model of each class. In this way, MCM-PHD-JTC has a more flexible modularized structure and facilitate the incorporation of extra models and extra classes, and the particles can be propagated according to their exact class-dependent kinematic model set thanks to the modularized structure. To achieve more robust and reliable performance, multi-sensor fusion is exploited. Demspter-Shafter (D-S) belief function is then incorporated into MCM-PHD-JTC under transferable belief model (TBM) to provide a flexible fusion result. Furthermore, the particle labeling method is introduced for track continuity, eventually addressing the joint tracking-association-identification-fusion problem in an integral framework efficiently. Moreover, because of no attribute sensors applied, the priori flight envelop information of targets is incorporated to provide classification. Simulations verify that the proposed multi-sensor multi-target MCM-PHD-JTC with TBM and track continuity shows reliable tracking and reasonable and correct classification with great flexibility.