Kernel-Based Target Tracking Using Bearing-Only Measurements

Passive infrared sensors serve as the primary sensing device for air-to-air missiles. However, the bearing-only measurements they provide often lead to challenges related to low observability and high nonlinearity, complicating the acquisition of complete state information. Furthermore, the targets being tracked frequently exhibit complex motion patterns due to aggressive maneuvers during evasion or combat, making it difficult to determine their motion models a priori. To tackle the tracking problem for the highly maneuvering targets in such complex scenarios, this paper presents a self-learning method based on data-driven Gaussian processes regression, which can learn nonlinear relationships from the data. In particular, a multi-kernel approach is employed to address the limitations of a single kernel, allowing for a more comprehensive capture of the diverse characteristics of the data and enhancing overall tracking performance. The effectiveness of the proposed multi-kernel Gaussian Process regression based tracking within the interacting multi-model framework is validated through simulations.