Ballistic coefficient estimation with Gaussian process particle filter

Abstract

This paper presents the Gaussian-process based particle filter for estimating the ballistic coefficient of a high-speed target. Tracking a high-speed target is a critical task in the estimation accuracy because any intervention on the target will require accurate information on the target states. While enumerating the target states, i.e. positions, velocities and acceleration, the states of the ballistic target will be affected by the ballistic coefficient. The interacting multiple model (IMM) is a dominant solution on the estimation of the coefficients, yet the selection and the updates of the coefficient hypotheses are difficult tasks. Hence, we adapt the Gaussian-process based particle filter for the hypotheses generations over time to enhance the performance of the IMM. The Gaussian process learns the over-time changes of the ballistic coefficient, so the next particle generation proposal can be better informed. Our experiments show a significant increase in the coefficient estimation accuracy as well as a consistent gain in the position estimation accuracy.