|T3.3.1 Intelligent AUV mapping|
|T3.3.2 AUV-ASV Cooperation|
Lead: Coronis, UdG, IQUA
M6-M19. A framework will be developed and integrated to detect and recognize potential interest targets while the AUV is operating at a relatively high exploratory altitude above the sea-floor. This exploratory altitude is defined as a trade- off between a large observation footprint to allow an efficient wide area survey, and having sufficient ground resolution to allow adequate target detection and recognition. The detection and recognition of targets will benefit from recent developments on acoustic and optical imaging analysis such as the application of Convolutional Neural Networks (CNN – Deep Learning) which allow real-time operation with modest computational hardware. The training of the CNN will be performed offline using a desktop Nvidia Titan V graphics card, while online processing once the CNN has been trained will be executed onboard using a Nvidia Jetson TX2. Methods will also be developed to define both the area around the target for the lower altitude survey and an adequate coverage pattern. This pattern will take into account the vehicle manoeuvrability constraints and define a safe profile for the survey altitude.