Team: Mike Allen, Dr Elena Gaura , Dr James Brusey.
In a physical and geometric context, localisation is the process of establishing the spatial relationships (i.e. relative positions) between devices or objects of interest. In wireless networked sensing, this can have several meanings, particularly self localisation and source localisation. The aim of the acoustic localisation project is to investigate both source and self localisation of acoustic signals within the context of wireless embedded sensing systems. This project targets resource constrained devices that communicate wirelessly and are equipped with acoustic sensing technologies (such as microphones).
Self localisation refers to all of the devices in the network establishing their relative (or global) position by whatever means they have available (e.g. acoustic, radio, laser, gps). The project focuses on in acoustic based self localisation, meaning nodes emit acoustic signals to one another in order to estimate distance (and potentially direction of arrival).
Source localisation comes from the signal processing field of array processing, and refers to the use of an array (in this case a network of localised wireless sensors) to estimate the position of an object of interest. With respect to acoustic signals, this can have application in gunshot localisation, smart fences (for border monitoring), speaker localisation (in smart offices) and animal localisation to name but a few. Motivating applications for this work are bioacoustics based, in which natural scientists want to detect, classify and localise animals in their natural habitat (specifically ant-birds and marmots), in order to understand more about their behaviour patterns.
Both self localisation and source localisation are important to give context; node positions must be well-known in order for the network to estimate the positions of any other unknown source relative to it. In general, is important to know spatially “where the data came from” in order to make correct observations about the phenomena being observed.
So far, this project has progressed in both self and source localisation directions. Work is ongoing to examine the effects of 3D terrain on sensor network acoustic based self-localisation algorithms. Typically, most existing self localisation algorithms make planar assumptions about network deployment that are not true in reality. Range estimation characterisation has been performed on several platforms of varying processing capability - the Mica2, Gumstix and Acoustic ENSBox, each resulting in vastly different operational ranges, accuracy and precision. This data will be used to drive simulations of the effects of 3D terrain of varying complexity on self localisation algorithms.
Source localisation has been addressed in a collaboration with UCLA and MIT, as part of the ongoing VoxNet project, which aims to develop hardware and software tools to aid data acquisition and processing in both on and off-line contexts for bioacoustics related applications. A system for on-line localisation of marmot calls has been developed, and techniques to minimise the time taken to gather acoustic event detections from nodes and process them into a position estimate. In the next stage, investigation into oversampling of regions with respect to position estimation accuracy will be explored, to help understand deployment densities for a given acoustics application.
For details of ongoing work go to Mike's PhD project page.