Underwater threat detection and tracking using multiple sensors and advanced processing

摘要

The vulnerability of military installations and critical infrastructure sites from underwater threats is now well accepted and, in order to combat these security weaknesses, there has been growing interest in - and adoption of - sonar technology. Greater availability of Autonomous/Unmanned Underwater Vehicles (A/UUVs) to both adversary nations and terrorists/saboteurs is also a cause of increasing concern. The small size and low acoustic target strength/signature of these vehicles presents significant challenges for sonar systems. The well-known challenges of the underwater environment, particularly in a harbor or port setting, can lead to a Nuisance Alarm Rate (NAR) that is higher than that of traditional security sensors (e.g. CCTV). This, in turn, can lead to a lack of confidence from end users and a possibility that `real' alerts are incorrectly dism issed. In the past this has been addressed by increasing the capability of individual sensors, leading to ever-increasing sensor complexity, however, the relationship between sensor performance and complexity/cost is highly non-linear. Even with the most complex and capable sensors, the fundamental limit to performance is often limited by acoustics, not sensor capability. In this paper we describe an alternative approach to reducing NAR and improving detection of difficult targets (e.g. UUVs), through intelligent combination and fusion of outputs from multiple sensors and data/signal processing algorithms. We describe the statistical basis for this approach, as well as techniques, methodologies and architectures for implementation. We describe the approach taken in our prototype algorithms/system, as well as quantitative and qualitative results from testing in a real-world environment. These results show a significant reduction in NAR and increase in classiflcation/alert range. Finally, we describe current focus areas for algorithmic and system development in both the short and medium term, as well as future extensions of these techniques to more classes of sensors, so that more challenging problems can be addressed.