Although underwater visual observation is an ideal method for detailed survey of seafloors, it is currently a costly process that requires the use of Remotely Operated Vehicles (ROVs) or Human Occupied Vehicles (HOVs), and can cover only a limited area. This paper proposes an innovative method to navigate an autonomous underwater vehicle (AUV) to create both 2D and 3D photo mosaics of seafloors with high positioning accuracy without using any vision-based matching. The vehicle finds vertical pole-like acoustic reflectors to use as positioning landmarks using a profiling sonar based on a SLAM (Simultaneous Localization And Mapping) technique. These reflectors can be either artificial or natural objects, and so the method can be applied to shallow vent fields where conventional acoustic positioning is difficult, since bubble plumes can also be used as landmarks as well as artificial reflectors. Path-planning is performed in real-time based on the positions and types of landmarks so as to navigate safely and stably using landmarks of different types (artificial reflector or bubble plume) found at arbitrary times and locations. Terrain tracker switches control reference between depth and altitude from the seafloor based on a local map of hazardous area created in real-time using onboard perceptual sensors, in order to follow rugged terrains at an altitude of 1 to 2 meters, as this range is ideal for visual observation. The method was implemented in the AUV Tri-Dog 1 and experiments were carried out at Tagiri vent field, Kagoshima Bay in Japan. The AUV succeeded in fully autonomous observation for more than 160 minutes to create a photo mosaic with an area larger than 600 square meters, which revealed the spatial distribution of detailed features such as tube-worm colonies, bubble plumes and bacteria mats. A fine bathymetry of the same area was also created using a light-section ranging system mounted on the vehicle. Finally a 3 D representation of the environment was created by merging the visual and bathymetry data.
The Proper Orthogonal Decomposition (POD) analysis method, which is one of principal component analysis techniques, was applied to time series tidal current vector data of Kagoshima Bay. Numerical simulation was applied to semidiurnal tidal conditions only. POD analysis has the advantage of revealing the dominant component with coherence and for visualizing as pattern recognition. The result of POD analysis for velocity component of the fluctuating current which didn't contain residual current components shows that the upper two modes were the dominant current structures. The eigen vector fields in modes 1 and 2 correspond to the characteristics of oscillating current and crossing current, respectively. However, the major axis of eigen vectors of each mode is not necessarily in agreement with the major axis of current ellipses. This is in accordance to lags of the tide of the original current data which shows the different values at some places. Therefore, the POD method has qualitative potential and needs further improvement for pre-quantitative evaluation.