Maritime Tracking of Past Data
L. D. Servi, (MIT Lincoln Laboratory), email@example.com
Maritime tracking, and maritime domain awareness more generally, has gained increasing interest in recent years as typified by a recommendation for improved systems in the Quadrennial Defense Review (QDR) published by the Department of Defense on February 6, 2006 (page 58). Maritime tracking problems can be segmented as real time or forensic depending on whether the track is estimated while the data is collected or after it is collected. This talk will concern only the latter case.
Fast forensic tracking, i.e., tracking a ship's previous path, could be useful in a number of applications in defense and national security. For example, the Coast Guard must approve all ships entering an US port and hence may be assisted by improved information about the ship's previous path. Alternatively, after a catastrophic event occurs at sea there is a need to forensically examine the ship(s) involved to assist in inferring attribution, finding the perpetrators, and/or identifying relevant perpetrator infrastructure related to the event. Finally there may be a desire to periodically update the location of suspicious but not imminently dangerous ships.
It is empirically useful to assume a ship's path consists of a concatenation of a number of paths parametrized by a small number of variables. In its simplest case, it could be idealized by a line segment or a piecewise linear paths.
This talk will describe insights into this forensic tracking problem first from the image processing literature point of view with a focus on Hough Transforms. Here the basic approach is to parametrize the space of the potential ship paths, construct a goodness of fit objective function, and then maximize the objective function. What makes this approach difficult is the largeness of the state space and the highly non-convex objective function found in realistic settings.
An alternative approach, which has been thus far been used only under the assumption of precise location measurements, will be presented and its performance illustrated using simulated data. In particular, it will be shown that the algorithm can identify a path consisting of 10 true measurements amid 1000 false measurements in .047 seconds and can identify a path consisting of 1000 true measurements amid 10,000 false measurements in .38 seconds (both on a dual 3 Ghz Xeon processor). For the case of imprecise location measurement, directions of future efforts will be summarized.
*This work was sponsored by the U.S. Government under Air Force Contract FA8721-05-C-0002. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the United States Government.