Trackers

class stonesoup.tracker.base.Tracker

Bases: Base

Tracker base class

class stonesoup.tracker.simple.SingleTargetTracker(initiator: Initiator, deleter: Deleter, detector: DetectionReader, data_associator: DataAssociator, updater: Updater)

Bases: _TrackerMixInNext, Tracker

A simple single target tracker.

Track a single object using Stone Soup components. The tracker works by first calling the data_associator with the active track, and then either updating the track state with the result of the updater if a detection is associated, or with the prediction if no detection is associated to the track. The track is then checked for deletion by the deleter, and if deleted the initiator is called to generate a new track. Similarly if no track is present (i.e. tracker is initialised or deleted in previous iteration), only the initiator is called.

Parameters:

• initiator (Initiator) – Initiator used to initialise the track.

• deleter (Deleter) – Deleter used to delete tracks.

• detector (DetectionReader) – Detector used to generate detection objects.

• data_associator (DataAssociator) – Association algorithm to pair predictions to detections

• updater (Updater) – Updater used to update the track object to the new state.

track

Current track being maintained. Also accessible as the sole item in tracks

Type:

Track

initiator: Initiator

Initiator used to initialise the track.

deleter: Deleter

Deleter used to delete tracks.

detector: DetectionReader

Detector used to generate detection objects.

data_associator: DataAssociator

Association algorithm to pair predictions to detections

updater: Updater

Updater used to update the track object to the new state.

update_tracker(time: datetime, detections: set[Detection]) → tuple[datetime, set[Track]]

Use time and detections to create tracks.

This is an alternative to iterating over the tracker, instead the tracker can be progressed by multiple calls to this method.

Parameters:

• time (datetime.datetime) – Datetime of current time step

• detections (set of Detection) – Detections existing in the time step to be tracked

Returns:

• datetime.datetime – Datetime of current time step

• set of Track – Tracks existing in the time step

class stonesoup.tracker.simple.SingleTargetMixtureTracker(initiator: Initiator, deleter: Deleter, detector: DetectionReader, data_associator: DataAssociator, updater: Updater)

Bases: _TrackerMixInNext, Tracker

A simple single target tracking that receives associations from a (Gaussian) Mixture associator.

Track single objects using Stone Soup components. The tracker works by first calling the data_associator with the active track, and then either updating the track state with the result of the data_associator that reduces the (Gaussian) Mixture of all possible track-detection associations, or with the prediction if no detection is associated to the track. The track is then checked for deletion by the deleter, and remaining unassociated detections are passed to the initiator to generate new track.

Parameters:

• initiator (Initiator) – Initiator used to initialise the track.

• deleter (Deleter) – Deleter used to delete tracks.

• detector (DetectionReader) – Detector used to generate detection objects.

• data_associator (DataAssociator) – Association algorithm to pair predictions to detections

• updater (Updater) – Updater used to update the track object to the new state.

initiator: Initiator

Initiator used to initialise the track.

deleter: Deleter

Deleter used to delete tracks.

detector: DetectionReader

Detector used to generate detection objects.

data_associator: DataAssociator

Association algorithm to pair predictions to detections

updater: Updater

Updater used to update the track object to the new state.

update_tracker(time: datetime, detections: set[Detection]) → tuple[datetime, set[Track]]

Use time and detections to create tracks.

This is an alternative to iterating over the tracker, instead the tracker can be progressed by multiple calls to this method.

Parameters:

• time (datetime.datetime) – Datetime of current time step

• detections (set of Detection) – Detections existing in the time step to be tracked

Returns:

• datetime.datetime – Datetime of current time step

• set of Track – Tracks existing in the time step

class stonesoup.tracker.simple.MultiTargetTracker(initiator: Initiator, deleter: Deleter, detector: DetectionReader, data_associator: DataAssociator, updater: Updater)

Bases: _TrackerMixInNext, Tracker

A simple multi target tracker.

Track multiple objects using Stone Soup components. The tracker works by first calling the data_associator with the active tracks, and then either updating the track state with the result of the updater if a detection is associated, or with the prediction if no detection is associated to the track. Tracks are then checked for deletion by the deleter, and remaining unassociated detections are passed to the initiator to generate new tracks.

Parameters:

• initiator (Initiator) – Initiator used to initialise the track.

• deleter (Deleter) – Deleter used to delete tracks.

• detector (DetectionReader) – Detector used to generate detection objects.

• data_associator (DataAssociator) – Association algorithm to pair predictions to detections

• updater (Updater) – Updater used to update the track object to the new state.

initiator: Initiator

Initiator used to initialise the track.

deleter: Deleter

Deleter used to delete tracks.

detector: DetectionReader

Detector used to generate detection objects.

data_associator: DataAssociator

Association algorithm to pair predictions to detections

updater: Updater

Updater used to update the track object to the new state.

update_tracker(time: datetime, detections: set[Detection]) → tuple[datetime, set[Track]]

Use time and detections to create tracks.

This is an alternative to iterating over the tracker, instead the tracker can be progressed by multiple calls to this method.

Parameters:

• time (datetime.datetime) – Datetime of current time step

• detections (set of Detection) – Detections existing in the time step to be tracked

Returns:

• datetime.datetime – Datetime of current time step

• set of Track – Tracks existing in the time step

class stonesoup.tracker.simple.MultiTargetMixtureTracker(initiator: Initiator, deleter: Deleter, detector: DetectionReader, data_associator: DataAssociator, updater: Updater)

Bases: _TrackerMixInNext, Tracker

A simple multi target tracker that receives associations from a (Gaussian) Mixture associator.

Track multiple objects using Stone Soup components. The tracker works by first calling the data_associator with the active tracks, and then either updating the track state with the result of the data_associator that reduces the (Gaussian) Mixture of all possible track-detection associations, or with the prediction if no detection is associated to the track. Tracks are then checked for deletion by the deleter, and remaining unassociated detections are passed to the initiator to generate new tracks.

Parameters:

• initiator (Initiator) – Initiator used to initialise the track.

• deleter (Deleter) – Deleter used to delete tracks.

• detector (DetectionReader) – Detector used to generate detection objects.

• data_associator (DataAssociator) – Association algorithm to pair predictions to detections

• updater (Updater) – Updater used to update the track object to the new state.

initiator: Initiator

Initiator used to initialise the track.

deleter: Deleter

Deleter used to delete tracks.

detector: DetectionReader

Detector used to generate detection objects.

data_associator: DataAssociator

Association algorithm to pair predictions to detections

updater: Updater

Updater used to update the track object to the new state.

update_tracker(time: datetime, detections: set[Detection]) → tuple[datetime, set[Track]]

Use time and detections to create tracks.

This is an alternative to iterating over the tracker, instead the tracker can be progressed by multiple calls to this method.

Parameters:

• time (datetime.datetime) – Datetime of current time step

• detections (set of Detection) – Detections existing in the time step to be tracked

Returns:

• datetime.datetime – Datetime of current time step

• set of Track – Tracks existing in the time step

Particle

class stonesoup.tracker.particle.SingleTargetExpectedLikelihoodParticleFilter(initiator: Initiator, deleter: Deleter, detector: DetectionReader, data_associator: DataAssociator, updater: ParticleUpdater, resampler: Resampler = None)

Bases: _BaseExpectedLikelihoodParticleFilter

An expected likelihood particle filter (ELPF) [1] for tracking a single target.

Track a single object using Stone Soup components. The tracker works by first calling the data_associator with the active track, and then either updating the track state with the result of the updater if a detection is associated, or with the prediction if no detection is associated to the track. The track is then checked for deletion by the deleter, and if deleted the initiator is called to generate a new track. Similarly, if no track is present (i.e. tracker is initialised or deleted in previous iteration), only the initiator is called.

Parameters:

• initiator (Initiator) – Initiator used to initialise the track.

• deleter (Deleter) – Deleter used to delete the track

• detector (DetectionReader) – Detector used to generate detection objects.

• data_associator (DataAssociator) – Association algorithm to pair predictions to detections

• updater (ParticleUpdater) – Updater used to update the tracks. It is important that no resampling is performed in the updater, as this is handled by the tracker.

• resampler (Resampler, optional) – Resampler used to resample the particles after the update step. If None, then SystematicResampler is used.

track

Current track being maintained. Also accessible as the sole item in tracks

Type:

Track

References

[1]

Marrs, A., Maskell, S., and Bar-Shalom, Y., “Expected likelihood for tracking in clutter with particle filters”, in Signal and Data Processing of Small Targets 2002, 2002, vol. 4728, pp. 230–239. doi:10.1117/12.478507.

class stonesoup.tracker.particle.MultiTargetExpectedLikelihoodParticleFilter(initiator: Initiator, deleter: Deleter, detector: DetectionReader, data_associator: DataAssociator, updater: ParticleUpdater, resampler: Resampler = None)

Bases: _BaseExpectedLikelihoodParticleFilter

An expected likelihood particle filter (ELPF) [2] for tracking multiple targets.

Track multiple objects using Stone Soup components. The tracker works by first calling the data_associator with the active tracks, and then either updating the track state with the result of the data_associator that reduces the (Particle) Mixture of all possible track-detection associations, or with the prediction if no detection is associated to the track. Tracks are then checked for deletion by the deleter, and remaining unassociated detections are passed to the initiator to generate new tracks.

Parameters:

• initiator (Initiator) – Initiator used to initialise the track.

• deleter (Deleter) – Deleter used to delete the track

• detector (DetectionReader) – Detector used to generate detection objects.

• data_associator (DataAssociator) – Association algorithm to pair predictions to detections

• updater (ParticleUpdater) – Updater used to update the tracks. It is important that no resampling is performed in the updater, as this is handled by the tracker.

• resampler (Resampler, optional) – Resampler used to resample the particles after the update step. If None, then SystematicResampler is used.

References

[2]

Marrs, A., Maskell, S., and Bar-Shalom, Y., “Expected likelihood for tracking in clutter with particle filters”, in Signal and Data Processing of Small Targets 2002, 2002, vol. 4728, pp. 230–239. doi:10.1117/12.478507.

Point Process

class stonesoup.tracker.pointprocess.PointProcessMultiTargetTracker(detector: DetectionReader, updater: Updater, hypothesiser: GaussianMixtureHypothesiser, reducer: GaussianMixtureReducer, extraction_threshold: Probability = 0.9, birth_component: TaggedWeightedGaussianState = None)

Bases: _TrackerMixInNext, Tracker

Base class for Gaussian Mixture (GM) style implementations of point process derived filters

Parameters:

• detector (DetectionReader) – Detector used to generate detection objects.

• updater (Updater) – Updater used to update the objects to their new state.

• hypothesiser (GaussianMixtureHypothesiser) – Association algorithm to pair predictions to detections

• reducer (GaussianMixtureReducer) – Reducer used to reduce the number of components in the mixture.

• extraction_threshold (Probability, optional) – Threshold to extract components from the mixture.

• birth_component (TaggedWeightedGaussianState, optional) – The birth component. The weight should be equal to the mean of the expected number of births per timestep (Poission distributed). The tag should be TaggedWeightedGaussianState.BIRTH

detector: DetectionReader

Detector used to generate detection objects.

updater: Updater

Updater used to update the objects to their new state.

hypothesiser: GaussianMixtureHypothesiser

Association algorithm to pair predictions to detections

reducer: GaussianMixtureReducer

Reducer used to reduce the number of components in the mixture.

extraction_threshold: Probability

Threshold to extract components from the mixture.

birth_component: TaggedWeightedGaussianState

The birth component. The weight should be equal to the mean of the expected number of births per timestep (Poission distributed). The tag should be TaggedWeightedGaussianState.BIRTH

update_tracks()

Updates the tracks (Track) associated with the filter.

Parameters:

self (GaussianMixtureMultiTargetTracker) – Current GM Multi Target Tracker at time \(k\)

Note

Each track shares a unique tag with its associated component

end_tracks()

Ends the tracks (Track) that do not have an associated component within the filter.

Parameters:

self (GaussianMixtureMultiTargetTracker) – Current GM Multi Target Tracker at time \(k\)

property extracted_target_states

Extract all target states from the Gaussian Mixture that are above an extraction threshold.

property estimated_number_of_targets

The number of hypothesised targets.

update_tracker(time: datetime, detections: set[Detection]) → tuple[datetime, set[Track]]

Use time and detections to create tracks.

This is an alternative to iterating over the tracker, instead the tracker can be progressed by multiple calls to this method.

Parameters:

• time (datetime.datetime) – Datetime of current time step

• detections (set of Detection) – Detections existing in the time step to be tracked

Returns:

• datetime.datetime – Datetime of current time step

• set of Track – Tracks existing in the time step