Novel Target Tracking Methods for Combining Passive and Active Sensors

Leonardo manufacture both active sensors and passive sensors for use with, for example, air platforms (e.g. F16s and drones). Active sensors, such as radars, can provide accurate range information but emit radiation to do so, reducing the extent to which sensing can be covert. Conversely, passive sensors, such as cameras, do not emit radiation in the same way, but are not easily configured to provide range information. This studentship relates to fusing the information from such sensors to maximise both the overall systems performance and the covert nature of the sensing.
Traditionally, the information from such sensors is combined at a high level of abstraction. However, the increasing availability of communications bandwidth and processing power make it possible to consider fusing the data at a lower level of abstraction. This should permit a cognitive approach to the fusion of information such that the sensors operation are adapted in real-time to explicitly maximise performance and covertness: for example, a radar might be used initially to infer an objects range but a camera might subsequently be used to maintain a track on the object. Such a cognitive capability is anticipated to improve the fidelity and accuracy of target tracking, in particular.
High performance target tracking algorithms, such as particle filters, are based on sequential Bayesian inference. This enables the algorithms to make best use of whatever information is available - such as measurements of position and speed, but also additional attributes such as colour. Such established algorithms can, at least in theory, cater with, for example, different individual sensors having different update rates, communication delays causing measurements to arrive out of order, sensor misalignments, systematic biases and using attributes to distinguish multiple objects tracks. As well as addressing these challenges, novel extensions to such algorithms will need to be developed to support the development of a cognitive capability. The studentship is therefore likely to draw on recent developments such as Sequential Monte Carlo (SMC) samplers; approximations commonly used in the context of Bayesian Networks, such as Structured Mean Field, Belief Propagation and Kikuchi approximations.
The PhD aims to develop these methods and to demonstrate performance in the context of a combination of inputs from multiple passive, active sensors and/or multi-function sensors. The objective is to enhance the accuracy and robustness of target tracking by both using advanced methods for processing the data but also by adapting the operation of the individual sensors such that they work synergistically to provide a cognitive capability.