|Source:||11th European Conference on Underwater Acoustics 2012, ECUA 2012, 2-6 July 2012, Edinburgh, PART 3, 34 1, 577-584|
Sonar · Defence, Safety and Security · Physics & Electronics · AS - Acoustics & Sonar · TS - Technical Sciences
A PHD particle filter implementation has been detailed for the fusion of measurements from multiple passive sonar nodes. It has been demonstrated on simulated metadata and on experimental passive acoustic data of divers and small boats collected in an operational port environment. Fusion at the metadata level lowers the necessary data bandwidth, which has practical benefits for long baseline systems. Moreover, the data fusion resolves sensor ambiguities permitting the localisation of targets, and results in an overall improvement in detection performance and a reduction of false alarms. The simulated and experimental results show great promise for our approach to data fusion. However, several improvements can be made and this will be one of the focuses of future work, including: - Improved implementation based on random finite sets (e.g., the Gaussian-mixture PHD filter for efficiency  and the cardinalised PHD filter for better performance ); - Improvements to the assumed sensor and target models and priors; - Inclusion of more measurements from the acoustic data (e.g., signature features) and possible integration with additional sensor types; - Self assessment of system performance and adaptation (e.g., adjusting the assumed clutter rate based on the conditions, such as during periods of rain or heavy shipping traffic).