A comparison of track initiation methods with the PMHT

Abstract

An important problem in multi-target tracking in clutter is that of track initiation and termination. When targets enter or leave the field of view of a sensor, the tracking algorithm must start new tracks or terminate existing tracks accordingly. One approach used to solve this problem is the addition of a visibility attribute to tracks. The visibility attribute provides a degree of filtering over the prior probability of measurement association and can be used as a statistic for track decisions. Standard model order estimation techniques suggest an alternative approach based on the log likelihood of the measurements. These approaches are used to derive initiation schemes for the probabilistic multi-hypothesis tracker (PMHT). The performance of these two approaches is compared with an intuitive ad hoc system through the use of receiver operating characteristic type curves.