Probabilistic Graphical Models in Multi-Target Tracking

This PhD is in the area of Statistical Signal Processing and Information Fusion, and is concerned with Joint Sensor Calibration and Hyperparameter Estimation in Multi-Target Tracking using Message Passing Algorithms.

Multi-target tracking (MTT) is an important problem in many defence and civilian applications, from tracking airborne targets, to maritime scenarios, to tracking people in urban environments. Recent work has been concerned with tracking an unknown number of targets from multiple multimodal asynchronous sensors, such as combining detections from multiple-radars and infra-red. Moreover, strong MoD investment in modular hierarchical autonomous sensor systems such as SAPIENT requires the development of inference algorithms for fixed-resource middleware platforms that scale with the variety, velocity, and veracity of the data, especially in cases where the sensor head has pre-processing capability.

However, multi-target tracking algorithms typically assume many hyperparameters that are often unknown but assumed, or heuristically determined, in existing algorithms presented in the literature. These include probability of detections and noise profiles, clutter profiles, sensor positions and orientations, and model dynamics. Recent work in the message-passing framework for MTT has attempted to incorporate joint estimation of these model parameters. However, current approaches still place a restriction on the model parameters in that they are drawn from a small finite discrete set, rather than the continuous-space in which they naturally reside.

This project will investigate message passing algorithms that explicitly deal with mixed continuous and discrete variables. In particular, we will develop relaxed structured decompositions of graphical models in mixed-variable problems, while considering scalable solutions, and consider recent message passing approximations. This includes realistically considering how many of these sensor and model hyperparameters can be estimated in linear time on a reasonable computational platform.

The University Defence Research Collaboration are pleased to invite applications for PhD studentships to work as part of a leading team of experts in signal processing and machine learning.

The project will be hosted by the Institute for Digital Communications at the School of Engineering at the University of Edinburgh and the student will work on the University Defence Research Collaboration (UDRC). The UDRC is a leading research partnership for signal processing for defence and develops new techniques to better transform data across many domains into actionable information, and meet the requirements for improved situational awareness, information superiority, and autonomy. This collaboration, sponsored by Dstl and the EPSRC, is academia-led and has commenced its third phase of research focusing on "Signal Processing in the Information Age". The Consortium is made up of the University of Edinburgh, Heriot-Watt University, Queen’s University Belfast and University of Strathclyde and there are currently PhD opportunities available across the four universities to work on diverse topics in signal processing, as part of a collaborative team. The work will involve strong links with industry and the UK defence sector. The PhD student will be expected to work closely with other research team members and to attend regular meetings to present project updates to the sponsors and partners of this project.

Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree.

Further information on English language requirements for EU/Overseas applicants can be found here - https://www.ed.ac.uk/studying/international/english/postgraduate