Alexander J. Pei
Target tracking is a critical component in defense and airspace protection. To provide awareness of potential enemy threats through target tracking, dynamic states are repeatedly updated based on observations. Because common dynamic models of moving objects typically use Cartesian coordinates, target tracking systems typically use this coordinate system as well. This presents a statistical challenge, however, when observations are recorded with different coordinate systems. This is the case with radar measurements, which use spherical coordinates (range, bearing, and elevation) instead of Cartesian coordinates (x, y, z). The main problem is integrating the statistics of new measurements with a priori state estimates to provide an updated a posteriori estimate. This article focuses on a converted-measurement approach to compute descriptive Cartesian statistics from spherical measurements for updates in a linear tracking system. Converted-measurement tracking, compared with mixed-coordinate tracking, can facilitate multisensor fusion in complex sensor networks. Various converted-measurement methods were evaluated, including Taylor approximations, unscented transforms, and debiased statistical methods, in a simple tracking scenario. Tracking performance varied across these three methods depending on the geometry of the scenario, so users of converted-measurement methods should evaluate the performance of each method for their given domain and application.
