Abstract
The need for harbor protection systems have increased over the last decade. One
vital component of harbor surveillance are the use of sonar to detect underwater
threats such as divers. In order to detect such threats, algorithms for detection,
tracking and a robust classi cation of underwater objects is needed.
This thesis uses known methods to detect, track and classify objects recorded from
real sonar data. A temporal cell averaging lter is used to detect objects in sonar
images and a tracking method based on the Probabilistic Data Association Filter
(PDAF) is used to track an object over time.
A set of object features, derived from a sequence of sonar images, is used to compute
a set of static and temporal features. The features tested in the thesis are compared
to each other to measure their ability to distinguish divers from marine life such as
seals and dolphins. A linear discriminant function is used as a classifer.