||Towards robust autonomous underwater target detection and identification using AUVs
||Thu 25 Mar 2010
||In recent years, Autonomous Underwater Vehicles have been increasingly used for Underwater MCM and asset protections missions. With the advent of robust platforms and the parallel introduction of high resolution imaging sensors such as synthetic aperture sonar and acoustic cameras, target detection and identification, as well as neutralisation are now possible in controlled environments and in simple scenarios. Multiple vehicle collaboration is also being investigated and initial results of collaborative behaviours are available.
However, there are still some outstanding key technological and scientific challenges to resolve for the technology to be widely usable and acceptable for non expert users:
• Goal based / adaptive mission planning. Currently platforms are really used as simple automatons realising a pre-programmed mission plan with little room for adaptation to the environment or the sensor performances. Missions are also programmed to a very low level of vehicle control which requires detailed knowledge of the platform capabilities. For real operations, it is likely that the initial mission will need to be adapted to respond to external (environment) and internal factors (faults). This is especially true for multi vehicle scenarios. User friendly goal-based planning is also a key element for the future acceptance of the systems.
• Efficient Data Mining of large data sets. Current target detection and classification methods are well suited to low to medium resolution sonar data processing but cannot cope with large data streams coming from high resolution SAS imagery, Acoustic and Video cameras. There is a need to review our current approaches to enable algorithms to quickly detect areas of interest and spend the computing resources wisely only focusing the expensive identification algorithms to small data streams.
• Reactive data gathering. It is unlikely that one sensor and one view of the data will provide sufficient information to enable effective classification in complex environments. Purposive vision techniques will enable to adapt the mission online in order to achieve specific PD/PFA objectives based on the sensor’s ability and current performances. The ability to manage multiple sensor modalities distributed on multi platforms is still an active research area that will be key to the success of such missions.
• World representation and sparse efficient data dissemination across multiple platforms. With high bandwidth sensors mounted on multiple platforms, distributed data fusion at symbolic level is a must, especially considering the low bandwidth of existing communication technology underwater.
In this paper, we will present the legacy work undertaken in the Oceans Systems Laboratory and SeeByte Ltd to solve some of the problems identified above. We will particularly focus on simulation tools, existing ATR algorithms and efficient schemes for data mining of large data sets and target identification algorithms. We will illustrate our technique on classical MCM and asset protection problems such as mine detection, identification and neutralisation and port and harbour surveillance.
<< Back to list of seminars