tracker = MultiTargetTracker(
initiator=initiator,
deleter=deleter,
detector=detector,
data_associator=data_associator,
updater=updater,
)
# %%
# Our tracker is built and our detections are ready to be read in from the CSV file, now we set the
# tracker to work. This is done by initiating a loop to generate tracks at each time interval.
# We'll keep a record of all tracks generated over time in a :class:`set` called `tracks`; as a
# :class:`set` we can simply update this with `current_tracks` at each timestep, not worrying about
# duplicates.
tracks = set()
for step, (time, current_tracks) in enumerate(tracker.tracks_gen(), 1):
tracks.update(current_tracks)
if not step % 10:
print("Step: {} Time: {}".format(step, time))
# %%
# Checking and plotting results
# -----------------------------
# The tracker has now run over the full data set and produced an output tracks. In this data set,
# from the below, we can see that we generated different number of tracks:
len(tracks)
# %%
# Versus the number of unique vessels/|MMSI|_:
len({track.metadata['MMSI'] for track in tracks})
# %%
# Once we've created a set of metrics, these are added to a Metric Manager, along with the
# associator. The associator can be used by multiple metric generators, only being run once as this
# can be a computationally expensive process; in this case, only SIAP Metrics requires it.
from stonesoup.metricgenerator.manager import SimpleManager
metric_manager = SimpleManager([basic_generator, ospa_generator, siap_generator, plot_generator],
associator=associator)
# %%
# Tracking and Generating Metrics
# -------------------------------
# With this basic tracker built and metrics ready, we'll now run the tracker, adding the sets of
# :class:`~.GroundTruthPath`, :class:`~.Detection` and :class:`~.Track` objects: to the metric
# manager.
for time, tracks in tracker.tracks_gen():
metric_manager.add_data(
groundtruth_sim.groundtruth_paths, tracks, detection_sim.detections,
overwrite=False, # Don't overwrite, instead add above as additional data
)
# %%
# With the tracker run and data in the metric manager, we'll now run the generate metrics method.
# This will also generate the plot, which will be rendered automatically below, which will give a
# visual overview
plt.rcParams["figure.figsize"] = (10, 8)
metrics = metric_manager.generate_metrics()
# %%
# So first we'll loop through the metrics and print out the basic metrics, which simply gives
# details on number of tracks versus targets.
deleter = UpdateTimeStepsDeleter(time_steps_since_update=2)
# Create a Kalman multi-target tracker
kalman_tracker = MultiTargetTracker(initiator=initiator,
deleter=deleter,
detector=sim,
data_associator=data_associator,
updater=updater)
# %%
# The final step is to iterate our tracker over the simulation:
kalman_tracks = {} # Store for plotting later
groundtruth_paths = {} # Store for plotting later
detections = [] # Store for plotting later
for time, ctracks in kalman_tracker.tracks_gen():
for track in ctracks:
loc = (track.state_vector[0], track.state_vector[2])
if track not in kalman_tracks:
kalman_tracks[track] = []
kalman_tracks[track].append(loc)
for truth in groundtruth_sim.current[1]:
loc = (truth.state_vector[0], truth.state_vector[2])
if truth not in groundtruth_paths:
groundtruth_paths[truth] = []
groundtruth_paths[truth].append(loc)
for detection in sim.detections:
detect_state = detection.measurement_model.inverse_function(detection)
loc = (detect_state[0], detect_state[2])