Track-to-truth metrics
- class stonesoup.metricgenerator.tracktotruthmetrics.SIAPMetrics(position_measure: Measure, velocity_measure: Measure)[source]
Bases:
MetricGenerator
SIAP Metrics
Computes the Single Integrated Air Picture (SIAP) metrics as defined by the Systems Engineering Task Force. The implementation provided here is derived from [1] and focuses on providing the SIAP attribute measures.
The SIAP metrics provided require provision of ground truth information.
In the original paper the calculations are dependent upon \(m\) which corresponds to the identifying number of the sense capability which is being assessed. This is not used in this implementation, with the assumption being that the fused sensor set is being assessed.
- Metrics:
Continuity (C): Fraction of true objects being tracked. The output is in the range \(0:1\), with a target score of 1.
Ambiguity (A): Number of tracks assigned to a true object. The output is unbounded with a range of \(0:\infty\). The target score is 1.
Spuriousness (S): Fraction of tracks that are unassigned to a true object. The output is in the range \(0:1\), with a target score of 0.
Positional Accuracy (PA): Positional error of associated tracks to their respective truths. The output is a distance measure, range \(0:\infty\), with a target score of 0.
Velocity Accuracy (VA): Velocity error of associated tracks to their respective truths. The output is a distance measure, range \(0:\infty\), with a target score of 0.
Rate of track number changes (R): SIAP continuity measure. Rate of number of track changes per truth. The output is in the range \(0:\infty\), with a target score of 0.
Longest track Segment (LS): SIAP continuity measure. Duration of longest associated track segment per truth. The output is a float (seconds), with a target score equal to the sum of all true object lifetimes.
- Reference
[1] Single Integrated Air Picture (SIAP) Metrics Implementation, Votruba et al, 29-10-2001
- Parameters:
- compute_metric(manager, **kwargs)[source]
Compute metrics:
\[\begin{split}\begin{alignat*}{3} \textrm{Name} &\quad \textrm{At Time} &&\quad \textrm{TimeRange}\\ C &\quad \frac{JT({t})}{J({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}}JT({t})} {\sum_{t_{start}}^{t_{end}}J({t})}\\ A &\quad \frac{N{A}({t})}{JT({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}} N{A}({t})}{\sum_{t_{start}}^{t_{end}}JT({t})}\\ S &\quad \frac{N({t}) - N{A}({t})}{N({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}} [N({t}) - N{A}({t})]}{\sum_{t_{start}}^{t_{end}}N({t})}\\ PA &\quad \frac{{\sum_{n\in tracks}PA_{n}(t)}}{NA(t)} &&\quad \frac{\sum_{t_{start}}^{t_{end}}{\sum_{n\in tracks}PA_{n}(t)}} {\sum_{t_{start}}^{t_{end}}{NA(t)}}\\ VA &\quad \frac{{\sum_{n\in tracks}VA_{n}(t)}}{NA(t)} &&\quad \frac{\sum_{t_{start}}^{t_{end}}{\sum_{n\in tracks}VA_{n}(t)}} {\sum_{t_{start}}^{t_{end}}{NA(t)}}\\ R &\quad -- &&\quad \frac{\sum_{j\in truths}NU_j-1}{\sum_{j\in truths}TT_j}\\ LS &\quad -- &&\quad \frac{\sum_{j\in truths}T{L}_{j}}{\sum_{j\in truths}T_{j}} \end{alignat*}\end{split}\]- Parameters:
manager (MetricManager) – containing the data to be used to create the metric(s)
- Returns:
Generated metrics
- Return type:
list of
Metric
objects
- static num_truths_at_time(manager, timestamp)[source]
\(J(t)\). Calculate the number of true objects held by manager at timestamp.
- Parameters:
manager (MetricManager) – Containing the data to be used
timestamp (datetime.datetime) – Timestamp at which to compute the value
- Returns:
Number of true objects held by manager at timestamp
- Return type:
- static num_associated_truths_at_time(manager, timestamp)[source]
\(JT(t)\). Calculate the number of associated true objects held by manager at timestamp.
- Parameters:
manager (MetricManager) – Containing the data to be used
timestamp (datetime.datetime) – Timestamp at which to compute the value
- Returns:
Number of associated true objects held by manager at timestamp
- Return type:
- static num_tracks_at_time(manager, timestamp)[source]
\(N(t)\). Calculate the number of tracks held by manager at timestamp.
- Parameters:
manager (MetricManager) – Containing the data to be used
timestamp (datetime.datetime) – Timestamp at which to compute the value
- Returns:
Number of tracks held by manager at timestamp
- Return type:
- static num_associated_tracks_at_time(manager, timestamp)[source]
\(NA(t)\). Calculate the number of associated tracks held by manager at timestamp.
- Parameters:
manager (MetricManager) – Containing the data to be used
timestamp (datetime.datetime) – Timestamp at which to compute the value
- Returns:
Number of associated tracks held by manager at timestamp.
- Return type:
- accuracy_at_time(manager, timestamp, measure)[source]
\(PA(t)\) or \(VA(t)\) (dependent on measure). Calculate the kinematic accuracy of track-truth associations held by manager at timestamp.
- Parameters:
manager (MetricManager) – Containing the data to be used
timestamp (datetime.datetime) – Timestamp at which to compute the value
measure (Measure) – Measure used to calculate ‘distance’ between truths and tracks
- Returns:
Kinematic accuracy of track-truth associations held by manager at timestamp.
- Return type:
Note
This method adds the ‘distance’ errors for each and every association. An alternative would be to consider each true object and track at most once.
- static truth_track_from_association(association)[source]
Find truth and track from an association.
- Parameters:
association (Association) – Association that contains truth and track as its objects
- Returns:
True object and track that are the objects of the association
- Return type:
- static total_time_tracked(manager, truth)[source]
\(TT\). Calculate the total time a truth is tracked for by tracks contained by manager.
- Parameters:
manager (MetricManager) – Containing the data to be used
truth (GroundTruthPath) – True object
- Returns:
Number of seconds that truth is tracked for
- Return type:
- static min_num_tracks_needed_to_track(manager, truth)[source]
\(NU_j\). Calculate the minimum number of tracks needed to track truth using tracks held by manager.
- Parameters:
manager (MetricManager) – Containing the data to be used
truth (GroundTruthPath) – True object
- Returns:
Minimum number of tracks needed to track truth
- Return type:
- rate_of_track_number_changes(manager)[source]
\(R\). Calculate the average rate of track number changes for true objects held by manager.
- Parameters:
manager (MetricManager) – Containing the data to be used
- Returns:
Average rate of track number changes
- Return type:
- static truth_lifetime(truth)[source]
\(T\). Calculate how long truth exists for.
- Parameters:
truth (GroundTruthPath) – True object
- Returns:
Number of seconds that truth exists for
- Return type:
- static longest_track_time_on_truth(manager, truth)[source]
\(TL_j\). Calculate the longest time that a single track is associated to truth using associations held by manager.
- Parameters:
manager (MetricManager) – Containing the data to be used
truth (GroundTruthPath) – True object
- Returns:
Number of seconds of longest association to truth
- Return type:
- class stonesoup.metricgenerator.tracktotruthmetrics.IDSIAPMetrics(position_measure: Measure, velocity_measure: Measure, truth_id: str, track_id: str)[source]
Bases:
SIAPMetrics
ID-based SIAP Metrics
SIAP metric generator that additionally computes ID-based SIAP metrics.
- ID-based Metrics:
ID Completeness (CID): ID-based SIAP. Fraction of true objects with an assigned ID. The output is in the range \(0:1\), with a target score of 1.
ID Correctness (IDC): ID-based SIAP. Fraction of true objects with correct ID assignment. The output is in the range \(0:1\), with a target score of 1.
ID Ambiguity (IDA): ID-based SIAP. Fraction of true objects with ambiguous ID assignment. The output is in the range \(0:1\), with a target score of 0.
Notes
This implementation assumes that track and ground truth path IDs are implemented via metadata, whereby the strings
track_id
andtruth_id
are keys to track and truth metadata entries with ID data respectively.Track
types store metadata outside of their states attribute. Therefore the ID SIAPs make metadata comparisons via the tracks’ last ID metadata values (as calling track.metadata will return a track’s metadata at the end of its life). To provide a better implementation, one might modifyTrack
types to contain a list of state types that hold their own metadata.
- Reference
[1] Single Integrated Air Picture (SIAP) Metrics Implementation, Votruba et al, 29-10-2001
- Parameters:
position_measure (
Measure
) – Distance measure used in calculating position accuracy scores.velocity_measure (
Measure
) – Distance measure used in calculating velocity accuracy scores.truth_id (
str
) – Metadata key for ID of each ground truth path in data-settrack_id (
str
) – Metadata key for ID of each track in data-set
- compute_metric(manager, **kwargs)[source]
Compute metrics:
\[\begin{split}\begin{alignat*}{3} \textrm{Name} &\quad \textrm{At Time} &&\quad \textrm{TimeRange}\\ C &\quad \frac{JT({t})}{J({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}}JT({t})} {\sum_{t_{start}}^{t_{end}}J({t})}\\ A &\quad \frac{N{A}({t})}{JT({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}} N{A}({t})}{\sum_{t_{start}}^{t_{end}}JT({t})}\\ S &\quad \frac{N({t}) - N{A}({t})}{N({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}} [N({t}) - N{A}({t})]}{\sum_{t_{start}}^{t_{end}}N({t})}\\ PA &\quad \frac{{\sum_{n\in tracks}PA_{n}(t)}}{NA(t)} &&\quad \frac{\sum_{t_{start}}^{t_{end}}{\sum_{n\in tracks}PA_{n}(t)}} {\sum_{t_{start}}^{t_{end}}{NA(t)}}\\ VA &\quad \frac{{\sum_{n\in tracks}VA_{n}(t)}}{NA(t)} &&\quad \frac{\sum_{t_{start}}^{t_{end}}{\sum_{n\in tracks}VA_{n}(t)}} {\sum_{t_{start}}^{t_{end}}{NA(t)}}\\ R &\quad -- &&\quad \frac{\sum_{j\in truths}NU_j-1}{\sum_{j\in truths}TT_j}\\ LS &\quad -- &&\quad \frac{\sum_{j\in truths}T{L}_{j}}{\sum_{j\in truths}T_{j}}\\ CID &\quad \frac{J{T}({t}) - J{U}({t})}{JT({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}}[J{T}({t}) - J{U}({T})]} {\sum_{t_{start}}^{t_{end}}J{T}({t})}\\ IDC &\quad \frac{J{C}({t})}{JT({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}} J{C}({t})}{\sum_{t_{start}}^{t_{end}}J{T}({t})}\\ IDA &\quad \frac{J{A}({t})}{JT({t})} &&\quad \frac{\sum_{t_{start}}^{t_{end}} J{A}({t})}{\sum_{t_{start}}^{t_{end}}J{T}({t})} \end{alignat*}\end{split}\]- Parameters:
manager (MetricManager) – containing the data to be used to create the metric(s)
- Returns:
Generated metrics
- Return type:
list of
Metric
objects
- find_track_id(track, timestamp)[source]
Find track ID at timestamp.
- Parameters:
track (Track) – Track object
timestamp (datetime.datetime) – Timestamp to retrieve ID data at
- Returns:
track ID at timestamp
- Return type:
any
- num_id_truths_at_time(manager, timestamp)[source]
- \(JU\), \(JC\), \(JI\). Calculate the number of true objects that are:
Un-identified, correctly identified, incorrectly identified at timestamp according to associations held by manager.
- Parameters:
manager (MetricManager) – Containing the data to be used
timestamp (datetime.datetime) – Timestamp at which to consider associations
- Returns:
Number of true objects: Un-identified, correctly identified, incorrectly identified
- Return type:
Note
A true object is considered to be un-identified if all tracks associated with it at timestamp have no ID.
A true object is considered to be correctly identified if all tracks associated with it at timestamp have the same ID as it.
A true object is considered to be incorrectly identified if all tracks associated with it at timestamp have a different ID to it.
A true object is considered to have ambiguous identification if tracks associated with it have differing ID (this value is calculated in the
compute_metric()
method).