Pattern metrics for groups of target tracks
Darin T. Dunham, Gregory J. Norgard, John D. Glass, Evan Everett, Dale Blair · 2018
The standard tracking metrics measure attributes such as position and velocity accuracy, covariance consistency, track completeness, spurious tracks, and redundant tracks. These standard metrics can be applied to a single target, and can be extended to a multiple-target scene by simply taking the per-target average. However, for quantifying tracking performance over a multiple-targets scene, standard per-target averages can be insufficient for characterizing scene-level performance. In this paper we present a family of pattern metrics which quantify how well the multiple-object track scene matches to truth at the scene level rather than at the per-target level. This paper will introduce and define three key pattern metrics: pattern accuracy, pattern consistency, and pattern containment. We will show that each of these pattern metrics provides a unique insight into system performance, and, to our knowledge, have not been previously published.