Tracking sperm whales using passive acoustics and particle filters.
University of Southampton, Institute of Sound and Vibration Research,
Passive acoustics provides a powerful tool for marine mammal research and mitigation of the risk posed by high energy anthropogenic acoustic activities through monitoring animal positions. Animal vocalisations can be detected and utilised in poor visibility conditions and while animals are dived. Marine mammal research is often conducted on restricted financial budgets by non-government organisations and academic institutions from boats or ships towing hydrophone arrays often comprising only two elements. The arrival time-delay of the acoustic wavefront from the vocalising animals across the array aperture is computed, often using freely available software, and typically regarded as the bearing of the animal to the array. This methodology is limited as it provides no ranging information and, until a boat manoeuvre is performed, whether the animal is to the left or right of the array remains ambiguous. Methods of determining range that have been suggested either negate the fact the animal is moving, rely on robust detection of acoustic reflections, rely on accurate equipment calibration and knowledge of the animal’s orientation or require modification of hydrophone equipment. There is a clear need to develop an improved method of estimating animal position as relative bearing, range and elevation to a hydrophone array or boat based on time-delay measurements. To avoid the costs of upgrading hydrophone arrays, and potentially the size of the vessels required to tow them, a software solution is desirable. This thesis proposes that the source location be modelled as a probability density function and that the source location is estimated as the mean. This is developed into a practical method using particle filters to track sperm whales. Sperm whales are the ideal subject species for this kind of development because the high sound pressure levels of their impulsive vocalisations (up to 236 dB re 1 ?Pa) makes them relatively simple to detect. Simulation tracking results demonstrate particle filters are capable of tracking a manoeuvring target using time-delay measurements. Tracking results for real data are presented and compared to the pseudotrack reconstructed from a tag equipped with accelerometers, magnetometers, a depth sensor and an acoustic recorder placed on the subject animal. For the majority of datasets the animal is tracked to a position relatively close to the surface sighting position. Sperm whales are typically encountered in groups, therefore a viable tracking solution needs to be capable of tracking multiple animals. A multiple hypothesis tracking method is proposed and tested for associating received vocalisations with animals, whereby vocalisations are correctly associated for periods exceeding 15 minutes
Actions (login required)