This project is concerned with improving the ability to detect faint objects in, for example, radar data. By improving detection performance, cheaper sensors will be able to emulate more expensive sensors, making the development of advanced detection algorithms very important in industrial settings. The project is co-created & co-funded by Leonardo, who make such sensors.
Conventional detectors apply a threshold to raw data that is collected over a sufficiently short period of time that the object hasn’t moved in a fully predictable way. Unfortunately, it can be challenging to set the threshold to generate detections from objects of interest and not also generate false alarms. Given the threshold, the resulting detections are typically then tracked (i.e., post-processed to “join the dots”) using statistical models for object movement.
Track-Before-Detect (TkBD) is an alternative approach that integrates the statistical models for object movement into the detector. This has the advantage that the threshold is applied to trajectories, not instantaneous positions and velocities: since object-generated data are different to noise on average (over the trajectory), this results in an improved ability to detect objects.
The model for the trajectory is important in defining the extent to which the object-generated data can be distinguished from noise-generated data. Perhaps surprisingly, existing TkBD make simplistic assumptions about objects’ trajectories (eg that they move in nearly-straight lines). This appears to limit the performance advantage achieved by TkBD to 6dB.
This project will investigate how to use statistical machine learning (eg techniques such as particle Markov chain Monte Carlo (PMCMC) or Sequential Monte Carlo (SMC) methods such as SMC2) to learn the parameters of non-linear models (eg models describing nearly-constant turns or models that capitalise on knowledge of the terrain). Such learning is computationally expensive and it is therefore strongly desirable to understand how such algorithms can capitalise on emerging many-cored computational resources (eg GPUs). It is anticipated that using a bank of class-specific non-linear models (each of which caters for a class of object that might be encountered) will increase the benefits that TkBD can achieve.
As well as extending the TkBD theory to cater for, for example, an unknown number of objects, the project will also investigate the extent to which many-cored architectures (particularly lower power/lower cost processors) can be capitalised upon to implement TkBD. The aim would be to develop a scalable processing architecture that allows large numbers of objects to be tracked across a distributed set of processors.
The key challenges are in developing real time processing methods for distributed processors that can use low-power processor systems and using adaptive scheduling to maintain energy efficiency across a number of processors. The non-academic partner (Leonardo) will be involved in defining the problem set and will help by supplying representative data for algorithm development and testing.
This project is part of the EPSRC Funded CDT in Distributed Algorithms: The What, How and where of Next-Generation Data Science. https://www.liverpool.ac.uk/research/research-themes/digital/cdt-distributed-algorithms/
The University of Liverpool is working in partnership with the STFC Hartree Centre and other industrial partners from the manufacturing, defence and security sectors to provide a 4 year innovative PhD training course that will equip over 60 students with the essential skills needed to become future leaders in data science, be it in academia or industry.
Every project within the centre is offered in collaboration with an Industrial partner who as well as providing co-supervision will also offer the unique opportunity for students to access state of the art computing platforms, work on real world problems, benchmarking and data. Our graduates will gain unparalleled experiences working across academic disciplines in highly sought-after topic areas, answering industry need.
As well as learning from academic and industrial world leaders, the centre has a dedicated programme of interdisciplinary research training including the opportunity to undertake modules at the global pinnacle of Data science teaching. A large number of events and training sessions are undertaken as a cohort of PhD students, allowing you to build personal and professional relationships that we hope will lead to research collaboration either now or in your future.
The learning nurtured at this centre will be based upon anticipation of the hardware recourses arriving on desks of students after they graduate, rather than the hardware available today.