New metrics for quantifying data association performance
Mark Silbert, Craig S. Agate · International Conference on Information Fusion · 2014
Numerous metrics exist for quantifying the performance of information fusion systems. Some metrics focus on estimation accuracy by comparing estimated quantities to the truth. Other metrics assess the accuracy of the estimation uncertainty by determining the consistency of the estimation error covariance. In this paper we define two metrics that quantify the data association algorithm's performance (whether the data are measurements or tracks). We compare the metric to a few existing metrics that quantify the effects of data association and evaluate the new metrics both with some notional examples and with some simulated data run through a track-to-track (T2T) fusion algorithm. Finally, we discuss a direct analogy between the data association problem and the information retrieval problem and reference two metrics in the information retrieval domain that are equivalent to the two metrics proposed in this paper.