Intended Audience: Researchers and Engineers working in the field of sensor networks and fusion applications.

Description: Networking multiple sensors can provide an enhanced picture of an environment compared to using individual sensors. Fusing information from these networked sensors, however, challenge the conventional approaches in terms of scalability and resource awareness.

Distributed processing paradigms aim to address these challenges and provide a scalable, robust and flexible computational structure by distributing the global tasks onto the network. Multisensor fusion algorithms, in a broad sense, aim to integrate the information exchanged between the nodes for enhancing the situation awareness.

Many multisensor fusion algorithms, however, rely on exact sensor calibration and do not account for errors in the parameters used to register the information from different sensors onto a common coordinate frame. Imperfect knowledge of these parameters can induce systematic errors and undermine the benefits of networked sensing. Therefore, calibration using sensor data is a very desirable capability.

This tutorial will present methods for fusion and registration in networks of sensors. In the first part, the focus will be on integrating information output by local filtering at the sensor nodes. Both optimal and suboptimal algorithms will be presented and discussed. The second part will cover registration/calibration of sensors. First, a centralised setting will be considered in which the sensor measurements are available at a centre. It will be shown how the registration process can exploit the Probability Hypothesis Density (PHD) filtering principles for handling the uncertainties in the multitarget model. The second topic will be a distributed setting in which several sensor nodes exchange filtered distributions as opposed to measurements. A recent solution will be introduced which feature local processing at the sensor nodes and message passing operations for selfcalibration.

Prerequisites: Target tracking, Bayesian filtering.

Presenter: Murat Uney, Simon Julier, and Daniel Clark

Murat Uney is a Research Fellow in the School of Engineering, University of Edinburgh. His research interests are in the broad scope of statistical signal and information processing with a particular emphasis on distributed, multimodal and resource constrained problem settings, and sensor fusion applications. He has industrial research and development experience in both aerospace/defence and telecommunication sectors. He gave a tutorial on distributed multisensor multiobject fusion algorithms in 2013 Summer School on Finite Set Statistics in Edinburgh. These algorithms were demonstrated online in a maritime environment, in collaboration with BAE Systems Advanced Technology Centre and University College London, and the results were presented in UDRC Summer School 2014 in Edinburgh. He covered optimal and adaptive filtering in the context of statistical signal processing in UDRC Summer School 2015, Surrey. His recent research focuses on sensor calibration in a distributed latent parameter estimation setting.

Simon Julier is a Reader in the Department of Computer Science at UCL. In 2014, he was a Distinguished Lecturer for the Aerospace & Electronic Systems Society. Before joining UCL, Dr Julier worked for nine years at the 3D Mixed and Virtual Environments Laboratory (3DMVEL) at the Naval Research Laboratory in Washington DC. From 2005 to 2007, he was the Principal Investigator of the ONRfunded Scalable Distributed Data Fusion Project ($700k) to develop techniques for robust fusion of multiple sources of data. Between 1999 and 2005 he was the Principal Investigator of the ONR funded Battlefield Augmented Reality System ($5.2M), a research effort to develop manwearable systems for providing situation awareness information. He focused on problems relating to information filtering, tracking and alignment, and error adaptation. He also served as the Associate Director of the 3DMVEL from 20052006. He was cochair of the IEEE VR 2006 and IEEE VR 2007 conferences. He received a DPhil in robotics from the Robotics Research Group, Oxford University, UK.

Daniel Clark is an Associate Professor in Sensors and Systems at HeriotWatt University. His research interests are in the development of the theory and applications of multiobject estimation algorithms for sensor fusion problems. He was chair of the 2013 Summer School on Finite Set Statistics in Edinburgh (with Dstl UDRC sponsorship) and Albuquerque (with AFOSR sponsorship).He has collaborated closely with Dstl in the UK on a number of projects in multitarget tracking spanning theoretical algorithm development to practical deployment in collaboration with BAE Systems, Finnmechanica, Thales, and DCNS. He lectures mathematics to undergraduate electrical engineers and developed a course on “MultiSensor Fusion and Tracking” for a European Masters programme (Vibot). In 2014, he was a Visiting Professor at the University of Colorado where he gave a lecture course on multiobject estimation. He gave a tutorial in 2011 at ICASSP with Branko Ristic entitled “Particle filters for multiobject Bayes filtering and sensor control in the framework of random set theory”.