gm_phd_tracker: Wrapper function to the Gaussian Mixture PHD tracker

In ick003/vesselett: A suite of target tracking algorithm, with a specialization for marine radar

Description

Wrapper function to the Gaussian Mixture PHD tracker

Usage

gm_phd_tracker(
  x,
  varnames,
  type = "constant-velocity-model",
  init_field = NULL,
  birth_field = NULL,
  mergingOption = "Panta2009",
  associationOption = "Panta2009",
  position_only = FALSE,
  T_s,
  sigma2_process,
  sigma2_measure,
  P_S,
  Jmax = 20,
  kappa,
  P_D,
  tau,
  U,
  wt,
  w,
  P_Sp,
  P_B,
  return.intensity = FALSE
)

Arguments

x	data.frame: requires time, utm position north and east (m), utm speed north and east (m/s).
varnames	vector: vector of names. Default to "north", "east", "v_north", "v_east".
type	string: random-walk, constant-velocity-model or coordinated-turn-model (Type of kinematic)
init_field	list: List of starting parameters
birth_field	list: List of birth parameters
mergingOption	string: How to merge the measures and create (or continue) the labeled tracks
associationOption	string: How to associate the new measures with the old tracks
position_only	logical: Position only observed (TRUE or FALSE)
T_s	scalar: Time interval to start the filter
sigma2_process	scalar, vector or matrix: Process variance of the kinematic movement
sigma2_measure	scalar, vector or matrix: Measurement variance
P_S	scalar: Probability of survival
Jmax	numeric: Maximum number of Gaussian mixtures carried over to the following time step
kappa	scalar: Intensity of clutter
P_D	scalar: Probability of detection
tau	scalar: Pruning threshold
U	scalar: Merging threshold
wt	scalar: Track confirmation threshold
w	scalar:
P_Sp	scalar: Probability of spawning
P_B	scalar: Probabilty of birth
return.intensity	logical: Returns all the parameters required to calculate the PHD (TRUE or FALSE)

Details

The GM PHD tracker that is implemented here proposes multiple options to choose from in terms of kinematic type (linear vs non-linear), data association options (through merging option and association optiob, defaulted both to Panta2009), and radar precision (position only vs position and velocity).

Value

data.frame with an additional column indication the target id, unique for the entire tracking history, and a few additional kinematic columns

References

\insertRef

Panta2009vesselett

Examples

## Not run: 
library(tidyverse)
library(ggrepel)
data(maritime2)
obs_x <- mutate(maritime2, time = time / 60, radar_north = (radar_north - min(north))/1000,
radar_east = (radar_east - min(east))/1000, radar_range = radar_range/1000,
north = (north - min(north))/1000, east = (east-min(east))/1000,
) %>% filter(between(time, 150,250))
plot_obs(obs_x, varnames = c("north", "east"))
mI = unlist(obs_x[obs_x$time == min(obs_x$time),c("north", "east", "v_north", "v_east")])
init_field = list(list(m = mI, P = diag(1, nrow = 4), w = 1, l = "1"))
res = gm_phd_tracker(obs_x, wt = 5e-1, sigma2_measure = 1e-2,
sigma2_process = 1e-1, kappa = 1e-10, tau = 1e-2, T_s = 1, check = FALSE,
init_field = init_field, P_B = 0.1, P_D = 0.9, P_S = 0.9)
Tracks = res$Tracks
Tracks <- left_join(Tracks,
unique(obs_x %>% select(time, radar_east, radar_north, radar_range)),
 by = "time") %>% arrange(time)
tracksP = Tracks %>%
select(north, east, time, target_id, radar_north, radar_east, radar_range) %>%
pivot_longer(c(north, east))
ggplot() +
    geom_point(data = tracksP ,
     aes(x = time, y = value, col = as.factor(target_id)), size = 1, shape = 3)+
    geom_line(data = tracksP ,
    aes(x = time, y = value, group = target_id),col = "black", size = 0.5) +
    facet_wrap( ~ name, ncol = 1, strip.position="left", scales = "free") +
    theme(panel.border = element_blank(),
    panel.grid.major = element_blank(),panel.grid.minor = element_blank(),
    panel.background = element_rect(fill = "white", colour = NA),
    legend.position = 'none',
    axis.line = element_line(size = 0.5, linetype = "solid", colour = "black"))

ggplot() + geom_point(data = obs_x %>% group_by(time, target_id) %>%
summarize(n = n()) %>%
ungroup() %>% group_by(time) %>% summarise(n =n()), aes(x = time, y = n)) +
geom_path(data = Tracks %>% group_by(time) %>%
summarize(n = n()), aes(x = time, y = n), col  ="red", size = 0.5) +
geom_path(data = Tracks %>% group_by(time) %>%
 summarize(n = round(sum(weight, na.rm=TRUE))), aes(x = time, y = n), col  ="blue", size = 0.5)+
geom_point(data = Tracks %>% group_by(time) %>%
 summarize(n = round(sum(weight, na.rm=TRUE))), aes(x = time, y = n), col  ="blue", size=1)

## End(Not run)

ick003/vesselett documentation built on July 20, 2020, 9:08 p.m.