Defence applications now routinely deploy a network of sensors (mobile or fixed) with heterogeneous and complementary sensing capabilities to achieve a particular set of goals. Advance in sensor technologies allows for a number of degrees of freedom in sensing devices (e.g. waveform selection and adaptation for radars) to be controlled by software, providing agility in real time operations. Exploiting a network of controllable sensors in an efficient way remains a challenging problem.
In this work package, we propose to study this sensor management problem in the context of a networked set of mobile sensors. We propose to approach its resolution with a coherent Bayesian framework modelling all the aspects of the problem, including:
- the development of observation models for each sensor, and the fusion of heterogeneous sensor data in a common observation process;
- the construction of a sensor policy that determines the optimal sensor allocations and configurations at each time step satisfying operational constraints (time, exclusion zones, energy) and following operational objectives (target detection and tracking, situational awareness)
E_WP5.1 Hierarchical sensor management for target tracking
The objective of this subpackage is to develop a coherent and unified framework integrating Bayesian estimation and sensor control in order to address sensor management problems for target tracking purposes. At the decision level, this subpackage will explore the integration of sensor decision policies within the Finite Set Statistics (FISST) methodology, which already provides a sound Bayesian framework for the detection and tracking of an unknown and time-varying number of targets with an arbitrary number of (possibly heterogeneous) sensors. At the sensor level, we will explore the development of observation models for detecting, characterizing and classifying targets from a variety of different sensors as well as adapting signalling to targets and local clutter (WP3) in order to produce sensor models compatible with the Bayesian framework.
E_WP 5.2 Computationally tractable solutions
These solutions will be developed using recent ideas in compressed sensing, finite set statistics (see WP3.1) and convex optimisation. Suboptimal solutions whose performance can be predicted and controlled will be investigated. How the problem space can be reduced by working at a higher level of abstraction in the DPDC loop (Direct-Capture-Process-Disseminate) whilst delegating the low-level sensor management to individual asset or groups of assets will be investigated.
E_WP 5.3 Multi-objective sensor management
This area will focus on:
So far the following points have been achieved:
Development of a novel information-based tool for the description of a target population, the higher-order regional statistics. Applied to multi-object filters, it helps the operator assessing the situational awareness, with a level of confidence, in any region of the surveillance space. In collaboration with WP2, it was implemented for two well-established multi-object filters in the tracking community, the Probability Hypothesis Density (PHD) and the Cardinalized PHD filters.
Design and implementation of the filter for Independent Stochastic Populations (ISP), that aims at providing a near-optimal solution to the multi-object Bayesian estimation problem with independent targets. This algorithm is the backbone solution exploited for the resolution of several multi-target tracking problems that spanning across our consortium: space situational awareness problems with very limited sensor coverage (collaboration with DSTL), closed-loop sensor management problems and multi-sensor maritime surveillance problems (collaboration with J. Franco, WP3, for a fast implementation of the ISP filter).
Production of performance metrics, adapted to the ISP filter, for decision making in the context of a closed-loop sensor management system Two solutions have already been designed: the higher-order regional statistics, and an alternative solution exploiting information metrics (e.g. the Rényi information measure).
- (1) Developing methods robust to network and assets failure
- (2) Taking into account multiple goals when the number of degrees of freedom in the system can provide redundancy that can be exploited to perform multiple objectives simultaneously.