Multi-target PHD Filter - (Vo et al., 2005)
Authors: Gwendoline De Bie - Johann Faouzi - Hicham Janati
Principal abstract object PHDabstract: serves as a root for the child objects (PHD) and (PHD_bootstrap). PHDabstract hides all plot and data generation methods that are commonly shared between the children classes.
Contains the (useful) PHD objects:
-
class PHD: general PHD filter with Importance Sampling density (q, q_pdf) mandatory in the prediction step where q is the random variable generator and q_pdf its density function.
-
Class PHD_bootstrap: PHD filter with IS density taken equal to the transition model given by (f,f_pdf) in the initiation. The separation is useful since with the bootstrap filter the prediction formulas are much simpler.
All parameters are stored:
Birth process properties, gaussian densities (for the moment) f and f_pdf, g and g_pdf (observation model) and their parameters.
-
Simulation.ipynb gives an example of a simulation with clutter parameter equal to r = 10 and r = 30.
-
The remaining notebooks compare the effect of number of particles per target and the choice of the state estimation method (Kmeans centroids or maximum weight).