Description of 3 Object Tracking Simulation


This simulation demonstrates multiple object tracking using particle filtering
techniques.  In this simulation, we have three dinghies (shown in the top-left
frame), each with a 7-dimensional state space: x- and y-location, orientation,
x- and y-velocities, change in orientation, and motion type.  Each dinghy has
three possible motion types: adrift, rowing, or motorized, and switches between
these three as a Markov chain.  Further, the dinghies interact by attracting
when far apart, and repelling when close together.

The observations (top-middle frame) which we get are similar to what we might
get when observing the ocean surface from above using an infrared camera.
Each pixel is corrupted by Gaussian noise, with a slightly higher mean
intensity of the pixel coincides with a dinghy's position.

At the bottom, we see the output of the filter.  The three frames represent
three possible numbers of targets.  There is a fourth possibility, that of
having no targets, which is not displayed.  The filter does not know,
beforehand, the correct number of targets.  The green bar at the bottom of each
of the frames represents the probability for each of the possible numbers of
targets.  Each frame shows an approximated density for the dinghies' positions
as a heat-scale, conditioned on the number of targets associated with that