Abstract: This dissertation addresses both theoretical and practical aspects
of target tracking in a distributed sensing environment. First we
consider the problem of tracking a target that moves according to a
Markov chain in a sensor network. We provide necessary and
sufficient conditions on the number of queries per time step to
track in three different scenarios: (1) the tracker is required to
know the exact location of the target at each time step; (2) the
tracker may lose track of the target at a given time step, but it is
able to ``catch-up," regaining up-to-date information about the
target's track; (3) tracking information is only known by the
tracker after a delay of d time steps. We then address the
inherent issues to multitarget tracking in urban terrains and the
development of techniques and algorithms that mitigate or eliminate
this issues. In order to improve the accuracy of track estimates
under such complex scenarios, it is important to use prior knowledge
of the environment. We investigate the integration of detection,
signal processing, tracking, and scheduling by exploiting distinct
levels of diversity: (1) spatial diversity through the use of
coordinated multistatic radars; (2) waveform diversity by adaptively
scheduling the transmitted radar waveform according to the scene
conditions; and (3) motion model diversity by using a bank of
parallel filters, each one matched to a different maneuvering model.
Finally, the resource allocation problem is considered. In
particular, we propose a two-level scheduling scheme that selects
waveforms and their transmitters for target tracking. The
lower-level scheduler selects waveforms to be transmitted at every
radar scan, and it is implemented as a round-robin or a myopic
algorithm for fast processing. At the top level, a non-myopic sensor
scheduler selects which transmitters should be active. While new
waveform scheduling decisions are taken at each radar scan, sensor
scheduling decisions remain in operation until the overall tracking
accuracy falls below a certain threshold, or a detection risk is
exceeded, defined by the number of consecutive scans a transmitter
is active.
Adviser: E. K. P. Chong Co-Adviser: Non-ECE Member: J. Hannig (Statistics) Member 3: J. R. Luo Addional Members:
