Inferring 3-dimensional animal motions from a set of 1-dimensional multibeam returns

Abstract

With the development of the FTV underwater multibeam sonar imaging system (Jaffe, 1995) and the more recent development of our combined optical and acoustical imaging system: OASIS (Jaffe, 1998), we now have the ability for both tracking and identifying animals (primarily zooplankton) in the sea. Past uses of the system have been mainly for abundance estimation, however, more recently we have become interested in attempting to infer the characteristics of animal behavior from the sonar returns. In the context of this problem, we sought to develop methods for processing the data which would be sensitive to extremely small motions of the animals, as it was suspected that the animals were mostly quiescent during some times of observation (daytime). As in most sonar systems, our system has much better range resolution that lateral or azimuth resolution. Tracking animals in three dimensions with this anisotropic resolution results in a coarser estimate of animal trajectory than desired. On the other hand, under the assumption of an isotropic distribution of animal motions the three dimensional probability density function for animal displacements: pdf3d(Δr′) should be derivable from a measurement of the one dimensional probability density function pdf1 d(Δρ) which measures displacements only in range. Since the one dimensional probability density function can be estimated from the measured variations in animal range with higher resolution than azimuth, using the methodology developed in this paper, one can obtain a higher accuracy version of the 3-dimensional pdf by using only the range information. Here, the methodology of the transformation is presented.KeywordsProbability Density FunctionTarget StrengthTarget DisplacementSonar SystemAzimuth ResolutionThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.