Distributed acoustic localization and tracking design and analysis.

摘要

Passive acoustic localization and tracking benefit the study of animal behavior by replacing or enhancing human involvement during field data collection. Collecting bio-acoustic data has traditionally been a difficult and time-consuming process in which researchers use portable microphones to record sounds while taking notes of their own detailed observations. Until recently, there has been no platform portable enough to develop such algorithms. The recent development of a portable and distributed acoustic sensing platform presents opportunities to develop automated tools for bio-acoustic field research. We propose the use of a sub-array (a node equipped with multiple sensors) as the basis node for the distributed platform and present the design and analysis for the optimal array geometry configuration. Having multiple sensors, each sub-array is capable of estimating the source direction-of-arrival (DOA) that can be combined by a fusion center into a position estimate. We propose using an approximate maximum likelihood (AML) method as the multi-source DOA estimator, whereby beamforming makes this approach suitable for source separation and signal enhancement. Analysis and simulation have shown that the AML method achieves the Cramer-Rao bound at a high signal-to-noise ratio. Furthermore, we demonstrate that a more computationally efficient algorithm exists that allows implementation for real time processing. A variant of the signal model has also allowed this algorithm to be extended for the reverberant scenario. The AML generated log-likelihood can also be fused directly producing a map for estimating the source position of a single source. When multiple sources are present, we propose using the random finite set (RFS) theory that allows randomly varying the number of sources and sensors to be modeled mathematically. The analogous Bayesian filtering generalized for RFS can automate localization by tracking the number of sources and their geo-kinematic information simultaneously. For practical implementation, we use particle filtering methodology to concentrate the search only in the relevant regions. The proposed localization system will be assessed from numerous controlled tests using pre-recorded source signals, a real bird monitoring at Chajul in Mexico, and a real marmot field monitoring scenario at Rocky Mountain Biological Laboratory in Colorado, USA. The proposed tracking system will be assessed using simulations.