An Operational Real-Time Large-Scale Visual Mosaicking and Navigation System

One use of ROVs (and potentially AUVs) is the generation of visual mosaics of areas of the ocean floor. A visual mosaic is formed by joining together multiple images taken by the vehicle as it surveys the sea floor to form a larger, composite view. In order to cover an area completely while guaranteeing overlap between images, precise measurement and control of the vehicle is required. External arrays such as LBLs can provide the precise navigation information required, at considerable expense in deployment and with limited operational range. Self-contained systems using DVLs and IMUS, do not suffer from these limitations but are subject to dead-reckoning drift. However, the imaging system itself can provide a means to eliminate drift. As long as the vision system can see a previously-visited area, the position measurement error relative to that part of the environment is bounded, and does not depend on vehicle path length or integration time. Additionally, the position information can be used to display a real-time mosaic to the user, which grows as the vehicle moves. Vision is subject to outages which do not affect a dead-reckoning sensor, such as dust clouds obscuring the view, and it is limited in its range from bottom. However, fusing the measurements of these two self-contained, complementary sensors - dead-reckoning and vision -provides a robust sensor which can provide the precision to guarantee complete area coverage and sufficient overlap in large-scale mosaics without the need to deploy an external positioning array. This paper presents an online seafloor mosaicking and navigation system which exploits the complementarity of dead-reckoning with a DVL and direct environment-relative sensing using vision. The mosaicking and navigation system operates in real time, and the calculated position provides a measurement which can be fed back to control the vehicle position relative to the environment and to display a navigation-grade mosaic to the user in real time. The results of field trials conducted in Monterey Bay using the MBARI ROVs Ventana and Tiburon are presented