Implementation considerations of the JPDA class of multitarget tracking algorithms

Multitarget tracking is a state estimation task wherein the trajectory of a real target is to be estimated (or tracked) in the presence of random clutter and of other nearby real targets which do not behave as random clutter. The joint probabilistic data association (JPDA) and joint probabilistic data association with possibly merged measurements (JPDAM) algorithms have been shown previously to provide the best tracking performance under these conditions. A key step in these algorithms is the calculation of the measurement-to-target associations, also known as the measurement-to-target weighting coefficients, and is the dominant computational load in the JPDA class of tracking algorithms. In this thesis we: advance a new computationally efficient method of calculating these coefficients called the "Fast JPDA" and the "Fast JPDAM" and specify a new analog optical architecture on which to realize the "Fast JPDA(M)". We also devise an error covariance matrix factor update algorithm for the fast and conventional JPDA(M) algorithms. And lastly, through simulations, we examine the wordlength required for the input measurements and the weighting coefficients and find that accuracy to 7 bits is sufficient for accurate tracking. We thus recommend our analog optical architecture and Fast JPDA(M) algorithms as integral parts of future multitarget tracking systems.