Previous   |   Session 154   |   Next  |   Author Index   |   Block Schedule

[154.03] A New Algorithm for Multiple Hypothesis-based Tracking and Discovery of Potentially Hazardous Near Earth Objects

A. Palaniappan (Yale University), J. N. Heasley (University of Hawaii), J. K. Uhlmann, K. Palaniappan (University of Missouri-Columbia)

The success of the Pan-STARRS NEO discovery program, will critically depend upon the automatic tracking of millions of observations, with the full sky covered on a monthly basis. Manual analysis and tracking of millions of NEO observations is not only rate limiting, but also highly error prone. Methods for fully automatic tracking of NEOs, which is similar to the missile-tracking problem, are being actively investigated. A new adjusted Kalman filter-based linear prediction and error model, is proposed for fast Mahalanobis-distance based gating using the KD-tree data structure, multiple hypothesis track pruning, and track assignment using a tree data structure. Our results using the novel (T,I) celestial coordinate system shows over 99% accuracy in correct pair-wise track associations for the same night observations, and remarkably over the longer four-night epochs, performance does not degrade as it does using other standard celestial coordinate systems including (\alpha,\delta), (\lambda,\beta), or (x,y,z). The (T,I) coordinates have a more linear behavior and so the Kalman predictions are more accurate. However, the Kalman prediction errors are biased and skewed. A new adjusted Kalman filter model is introduced to recursively incorporate the actual prediction error statistics using the test data. The multiple hypothesis tree containing all feasible tracks at the final epoch of tracking has over a 90% success rate in finding the correct asteroid associations within the top 30 tracks as ranked in terms of the average along track Mahalanobis distance. Future work includes incorporating joint assignment to prune the tree, and the use of a non-linear elliptical orbit prediction model combined with Monte Carlo covariance estimation methods to eliminate false tracks from the final set of long associations. Both of these extensions should further improve performance towards achieving automatic multi-target tracking of millions of asteroids.

Previous   |   Session 154   |   Next

Bulletin of the American Astronomical Society, 37 #4
© 2005. The American Astronomical Soceity.