Segmentation of images from an acoustic lens sonar

The Dual-Frequency Identification Sonar (DIDSON) is a high frequency (1.0 or 1.8 MHz), high resolution underwater imaging device. An important stage in the classification and identification of objects resting on the seabed is the estimation of object height and width. These estimates can be made based on the acoustic shadow cast by such objects when illuminated by high frequency active sonar pings. Several data sets have been obtained with a DIDSON mounted on the front of an autonomous undersea vehicle in a forward looking configuration. These data sets contain multiple views of some bottom-resting targets. The pre-processing steps necessary for segmentation of DIDSON images are described. The data are then analyzed, concentrating on the segmentation of shadow regions for follow-on estimation of target dimensions. Segmentation algorithms often depend on assumptions about the statistics of shadow and non-shadow regions in the images. These statistics are examined for other collected data sets, and their effect on segmentation algorithms is discussed