Comparison of Data Fusion Techniques for Target Detection with a Wide AzimuthSonar

摘要

A group at the Charles Stark Draper Laboratory developed a concept for a minereconnaissance platform called Intelligent Sonobuoy. This platform utilizes a low frequency sonar with wide aspect angle coverage. Furthermore the platform is designed to drift past an area of interest and thus obtain multiple detections from each sonar target. This thesis examines methods of fusing together those detections into a composite map of the target field in order to detect and localize those sonar targets. A technique based on hypothesis testing and maximum likelihood estimation is first derived and then applied to simulated data. Lastly, the system is validated on actual test data obtained in Mendum's Pond, New Hampshire during the summer and Fall of 1994. This system is shown to be effective at resolving targets to within a few meters. A competing approach based on the Hough transform is next examined. This clustering technique is applied to find the change in target location with respect to the buoy's position. The system works for simulated test data with a small number of detections. System performance declines rapidly as the number of detections increases and the system does not work well with the actual test data. (KAR) p. 4.