The pmht with hysteresis
Samuel J. Davey, Douglas Andrew Gray · 2003
The Probabilistic Multi-Hypothesis Tracker (PMHT) is a recent algorithm based on the application of Expectation Maximisation to multi-target tracking. The al- gorithm treats the sensor measurements as independent ob- servations of a mixture model, and the assignment of each measurement as an independent realisation of the mixing process. A result of this is that data association benveen tracks and measurements is performed independently for each measuremenr. If the miring proportions of the mod- els are unknown. then the PMHT estimates them, assuming them to be time independent. For most applications, the mixing proportions will not be time independent, but rather change smoothly with time. The PMHT with Hysteresis pro- vides a state model for the mixing proportions, allowing the PMHT to model them as random processes with arbitrary statistics.