Tracker Operating Characteristic for Integrated Probabilistic Data Association

Abstract

Integrated Probabilistic Data Association (IPDA) is a popular approach to target tracking and to track initiation and termination. In regards to track initiation, it may be said to belong to the "detect-then-track" class in which processing begins with sensor-level detections. This is in contrast to "track-before-detect" (TkBD) approaches in which un-thresholded sensor-level measurements are directly incorporated into a track-initiation statistic. In IPDA, the track-initiation statistic is the probability of existence (PoE). How should the confirmation threshold be set? We perform a statistical analysis that yields an accurate approximation of the false-track and track detection probabilities as a function of the threshold on the PoE. From these probabilities, IPDA's tracker operating characteristic (TOC) can be derived. We further propose a tracker using similar ideas but which belongs to the TkBD class. Its TOC is likewise derived and compared with IPDA. The results of numerical simulations are presented which support the theoretical derivations.