Tuesday, September 19, 2017
      CMRE Facebook page  CMRE LinkedIn page  CMRE PAO Youtube page
   
Text Size
CMRE banner

GPU technology advances SAS real-time applications in mine hunting

20130111 MUSCLESASseafloor 400

CMRE successfully exploits near real-time processing of Synthetic Aperture Sonar data for adaptive track spacing and target detection on an autonomous vehicle.

The CMRE "Autonomous Mine Search Using High-Frequency Synthetic Aperture Sonar" project focuses on increasing the efficiency and effectiveness of the full range of mine search missions. This is accomplished in part by greatly improving the processing and decision-making capabilities on-board the MUSCLE (Minehunting UUV for Shallow water Covert Littoral Expeditions) Autonomous Underwater Vehicle, which uses multi-resolution, multi-aspect Synthetic Aperture Sonar to create detailed images of the seafloor.

GPUs and their development tools like the CUDA libraries have been a key component for the success of recent sea trials where the full sonar processing chain was executed on the autonomous vehicle during the mission using a special embedded version of the NVIDIA Tesla GPU named GT240.

During the Autonomous Reactive Intelligence Sea Experiment (ARISE) '12 sea trial, in particular, the technology has been successfully tested under realistic mine hunting conditions. Experiments with a HUGIN AUV with HISAS 1030 SAS have also been conducted as a result of collaboration with Kongsberg Maritime and the Norwegian Defence Research Establishment (FFI).

The first objective of the Project uses techniques from machine intelligence to develop AUV behaviours that can help ensure complete coverage of a mine hunting area. For example, when sonar is used in areas with a rippled seabed, mine detection can be compromised at certain look angles with respect to the developed ripple field. Scientists are programming a behaviour so that the AUV can detect this type of seabed and then move into a better position.
The second objective is to develop automatic target recognition (ATR) algorithms that can increase confidence that an object is classified as a mine. Accurate classification is key to efficient autonomous mine countermeasures. Moving the algorithms and behaviors on-board the AUV has the potential to significantly increase the speed of operation for mine countermeasure missions.

The key to fulfill these two objectives is the capability to process the sonar data on-board the vehicle in near real-time, producing detailed acoustic images of the seabed. These images are then analyzed on the fly by the vehicle's computers, and specialized algorithms can then cause the AUV to take fully autonomous decisions on the best way to continue the mission in order to meet the specific objectives.

Various Synthetic Aperture Sonar (SAS) processing algorithms had been developed over several years at CMRE, existing in various implementations, although none of them were ideally suited for an embedded and near real-time application. Indeed, GPUs and the CUDA libraries seemed an ideal candidate for the task allowing for high scalability and code reusability. With support of the Istituto Italiano di Tecnologia (IIT), the entire processing chain was re-designed and implemented in C++ and CUDA. Extensive tests were conducted on mid- to high-end workstations equipped with different generations of TESLA GPUs, exhibiting speed-ups in excess of 70X from the original un-optimized scientific code. In parallel with that effort, the Engineering Division at CMRE started the integration of a new payload section for the MUSCLE AUV containing a ruggedized CPU and a ruggedized version of a TESLA GT240 GPU.

The new version of the CMRE SAS processing algorithms is now capable of running in near real-time on the NVIDIA GPU installed into the MUSCLE AUV. That's a huge step forward from the basic approach where the AUV performs missions following pre-programmed paths collecting raw sensor data to be analyzed at the end of the survey, which results in no capability to adapt to unexpected environmental conditions and/or sonar performance.

As usual at CMRE, the tests involved a multinational team of scientists and engineers aimed at finding the best technologies that could be used by Nations in joint NATO mine countermeasures missions.

"Autonomous Mine Search Using High-Frequency Synthetic Aperture Sonar" is one of the main projects at the Centre, aiming at increasing the capabilities of autonomous underwater vehicles (AUVs) by using Synthetic Aperture Sonar (SAS) to quickly and reliably detect, classify and localize mines. This implies the use of high performance, efficient processing systems installed into the underwater vehicles.

The CMRE focus is now on the integration of the newest generation of TESLA GPUs to allow for a larger portion of the entire processing chain (including ATR and autonomy algorithms) to run on the GPU.

User Login