Image-based motion estimation of underwater tow body

Abstract

During deep sea exploration, the motion of the deep tow system is influenced by that of the surface vessel, the change of the cable length and the current field in the water column. Though the towing vessel moves in a straight line on the sea surface, the heading of the tow body is not necessary collinear to the towing direction. If no devices are available to adjust the direction of the tow body, data collected by directional sensors, like camera or side scan sonar, will be skewed. It causes problems and difficulties in the post processing for mosaicing data sets into a map of a larger area. Moreover, the existence of the offset angle creates a staircase-like pattern on the map when the images are stitched together. The staircase pattern hampers the continuity of the data. The offset angle, in theory, can be computed from the USBL and the compass readings. But in practice, the refresh rates of USBL and the compass are not compatible such that the results are not good enough for maneuvering the tow system. In this paper, we propose using the feature points on the images acquired to estimate the offset angle of the tow system with respect to the trace of the motion. The computation is carried out with the OPENCV library implemented in Visual C. Currently, the refresh rate can be about 2 Hz with 30 fps camera acquisition, given the tow body is moving with 0.5 m/s speed.