Pattern classifier and method for associating tracks from different sensors

• US 20060279453A1
• Filed: 06/13/2005
• Published: 12/14/2006
• Est. Priority Date: 06/13/2005
• Status: Active Grant

• Current Assignee: Raytheon Company (Rtx Corporation)
• Original Assignee: Raytheon Company (Rtx Corporation)
• Inventors: Caber, Timothy E.
• Granted Patent: US 7,236,121 B2
• Time in Patent Office: Days
• US Class Current: 342/97
• CPC Class Codes: G01S 13/726 Multiple target tracking

Abstract

An interceptor-based sensor clusters tracks of objects to generate track clusters based on an uncertainty associated with each track, and generates feature vectors for a cluster under test using the relative placement and the population of other track clusters. The feature vectors may include one or more of a cluster count feature vector (N), a cluster population density feature vector (P), a cluster proximity feature vector (r), a cluster-weighted centroid feature vector (L) and a cluster scattering feature vector (θ). The interceptor-based sensor generates belief functions (μ) from corresponding feature vectors of clusters of tracks generated from a ground-based sensor and the interceptor-based sensor. The interceptor-based sensor may also associate the tracks with a cluster having a track of interest identified by a ground-based sensor based on the belief functions and may select one of the tracks for intercept of a corresponding object within the threat object cloud.

Claims

1. An interceptor-based sensor comprising:

a track clustering element to cluster tracks of objects to generate track clusters based on an uncertainty associated with each track;

a feature generating element to generate feature vectors for the track clusters in one or more directions with respect to a cluster under test, the feature vectors comprising one or more of a cluster count feature vector (N), a cluster population density feature vector (P), a cluster proximity feature vector (r), a cluster-weighted centroid feature vector (L) and a cluster scattering feature vector (θ

); and

a track selection element to associate tracks provided by another sensor based on belief functions generated from the feature vectors, the track selection element to select one of the tracks corresponding to a track of interest.

2. The sensor of claim 1 wherein the track selection element is to generate the belief functions (μ

) from corresponding feature vectors of both sensors for each of a plurality of cluster as clusters-under-test, and

wherein the track selection element is to assign an association class based on the belief functions and a plurality of ratings for selection of a track cluster having a high-probability of including the track of interest.

3. The sensor of claim 1 further comprising a track state estimation element to generate a track-state vector for each of plurality of the objects detected within a threat object cloud, the objects being detected by on-board sensor circuitry of the sensor, each track-state vector comprising velocity, position and the uncertainty,

wherein the track clustering element is to cluster the tracks based associated track-state vectors having overlapping uncertainty regions, the uncertainty regions based on the covariance associated with each track-state vector.

4. The sensor of claim 3 wherein the other sensor is a ground-based tracking sensor, and wherein radar tracks of a plurality of radar-tracked objects are received from the ground based sensor, and

wherein one of the radar-tracks is designated as the track of interest by the ground-based tracking sensor,

wherein the track clustering element is to further cluster the radar-tracks of the radar-tracked objects having overlapping uncertainty regions with the track of interest to generate a track-cluster of interest, and

wherein the feature generating element is to further generate feature vectors for the track-cluster of interest in two or more dimensional directions with respect to the track of interest.

5. The sensor of claim 4 further comprising coordinate transform processing element to transform a multi-dimensional threat-object map (TOM) provided by the ground-based tracking sensor to a two-dimensional threat-object map (TOM) comprising the radar tracks, wherein each radar-track comprises at least velocity, position and uncertainty information.

6. The sensor of claim 2 wherein the feature generating element computes the cluster count feature vector (N) by summing a number of the track clusters in each of a plurality of two or more dimensional directions with respect to the cluster under test, and

wherein the feature generating element generates a rating corresponding to either few or many based on a value of the cluster count feature vector (N) in a particular direction.

7. The sensor of claim 2 wherein the feature generating element computes the cluster population density feature vector (P) based on dividing a sum of a number of track clusters in a direction by a total number of track-clusters, and

wherein the feature generating element generates a rating corresponding to either sparse or dense based on the value of the cluster population density feature vector (P) in a particular direction.

8. The sensor of claim 2 wherein the feature generating element computes the cluster proximity feature vector (r) based on a population-weighted radial distance to the track clusters for a particular direction, and

wherein the feature generating element generates a rating corresponding to either urban or suburban based on a value of the cluster proximity feature vector (r) in a particular direction.

9. The sensor of claim 2 wherein the feature generating element computes the cluster-weighted centroid feature vector (L) based on a population-weighted mean in a particular direction divided by a maximal scene extent distance, and

wherein the feature generating element generates a rating corresponding to either near or far based on a value of the cluster-weighted centroid feature vector (L) in a particular direction.

10. The sensor of claim 2 wherein the feature generating element computes the cluster scattering feature vector (θ

) based on an angular deviation of the clusters with respect to a particular direction, and

wherein the feature generating element generates a rating corresponding to either scattered or aligned based on a value of the cluster scattering feature vector (θ

) in the particular direction.

11. The sensor of claim 1 wherein the track clustering element, the feature generating element, and the track selection element operate on one or more processors.

12. The sensor of claim 1 wherein the track selection element is to select the track of interest for intercept of a corresponding object within a threat object cloud.

13. A method of associating tracks from different sensors comprising:

clustering tracks of objects to generate track clusters based on an uncertainty associated with each track;

generating feature vectors for the track clusters in one or more directions with respect to a cluster under test, the feature vectors comprising one or more of a cluster count feature vector (N), a cluster population density feature vector (P), a cluster proximity feature vector (r), a cluster-weighted centroid feature vector (L) and a cluster scattering feature vector (θ

);

associating tracks provided by another sensor based on belief functions generated from the feature vectors; and

selecting one of the tracks corresponding to a track of interest.

14. The method of claim 13 further comprising:

generating the belief functions (μ

) from corresponding feature vectors of both sensors for each of a plurality of cluster as clusters-under-test; and

assigning an association class based on the belief functions and a plurality of ratings for selection of a track cluster having a high-probability of including the track of interest.

15. The method of claim 13 further comprising:

generating a track-state vector for each of plurality of the objects detected within a threat object cloud, the objects being detected by on-board sensor circuitry of the sensor, each track-state vector comprising velocity, position and the uncertainty, and

wherein clustering comprises clustering the tracks based associated track-state vectors having overlapping uncertainty regions, the uncertainty regions based on the covariance associated with each track-state vector.

16. The method of claim 15 wherein the other sensor is a ground-based tracking sensor, and wherein radar tracks of a plurality of radar-tracked objects are received from the ground based sensor,

wherein one of the radar-tracks is designated as the track of interest by the ground-based tracking sensor, and

wherein the method further comprises:

clustering the radar-tracks of the radar-tracked objects having overlapping uncertainty regions with the track of interest to generate a track-cluster of interest; and

generating feature vectors for the track-cluster of interest in two or more dimensional directions with respect to the track of interest.

17. The method of claim 16 further comprising transforming a multi-dimensional threat-object map (TOM) provided by the ground-based tracking sensor to a two-dimensional threat-object map (TOM) comprising the radar tracks, wherein each radar-track comprises at least velocity, position and uncertainty information.

18. The method of claim 14 further comprising:

computing the cluster count feature vector (N) by summing a number of the track clusters in each of a plurality of two or more dimensional directions with respect to the cluster under test; and

generating a rating corresponding to either few or many based on a value of the cluster count feature vector (N) in a particular direction.

19. The method of claim 14 further comprising:

computing the cluster population density feature vector (P) based on dividing a sum of a number of track clusters in a direction by a total number of track-clusters; and

generating a rating corresponding to either sparse or dense based on the value of the cluster population density feature vector (P) in a particular direction.

20. The method of claim 14 further comprising:

computing the cluster proximity feature vector (r) based on a population-weighted radial distance to the track clusters for a particular direction; and

generating a rating corresponding to either urban or suburban based on a value of the cluster proximity feature vector (r) in a particular direction.

21. The method of claim 14 further comprising:

computing the cluster-weighted centroid feature vector (L) based on a population-weighted mean in a particular direction divided by a maximal scene extent distance; and

generating a rating corresponding to either near or far based on a value of the cluster-weighted centroid feature vector (L) in a particular direction.

22. The method of claim 14 further comprising:

computing the cluster scattering feature vector (θ

) based on an angular deviation of the clusters with respect to a particular direction; and

generating a rating corresponding to either scattered or aligned based on a value of the cluster scattering feature vector (θ

) in the particular direction.

23. The method of claim 13 wherein the clustering, the feature generating and the track selecting are performed by one or more processors.

24. The method of claim 13 wherein selecting comprises selecting the track of interest for intercept of a corresponding object within a threat object cloud.

25. A pattern classifier comprising:

a track clustering element to cluster tracks of objects provided by a first sensor to generate track clusters based on an uncertainty associated with each track;

a feature generating element to generate feature vectors for the track clusters in one or more directions with respect to a cluster under test, the feature vectors comprising one or more of a cluster count feature vector (N), a cluster population density feature vector (P), a cluster proximity feature vector (r), a cluster-weighted centroid feature vector (L) and a cluster scattering feature vector (θ

); and

a track selection element to associate tracks provided by a second sensor (204) based on belief functions generated from the feature vectors, the track selection element to select one of the tracks corresponding to a track of interest.

26. The pattern classifier of claim 25 wherein the track selection element is to generate the belief functions (μ

) from corresponding feature vectors of both sensors for each of a plurality of cluster as clusters-under-test, and

wherein the track selection element is to assign an association class based on the belief functions and a plurality of ratings for selection of a track cluster having a high-probability of including the track of interest.

27. The pattern classifier of claim 25 further comprising a track state estimation element to generate a track-state vector for each of plurality of the objects detected by the first sensor, each track-state vector comprising velocity, position and the uncertainty,

wherein the track clustering element is to cluster the tracks based associated track-state vectors having overlapping uncertainty regions, the uncertainty regions based on the covariance associated with each track-state vector.

28. A missile-defense system comprising:

a ground-based sensor to acquire a threat cloud comprising a missile from a track-state estimate and covariance provided by an overhead sensor; and

an interceptor to receive track-state vectors of objects in the threat cloud tracked by the ground-based sensor,

the interceptor comprising a track clustering element to cluster tracks of objects to generate track clusters based on an uncertainty associated with each track, a feature generating element to generate feature vectors for the track clusters in one or more directions with respect to a cluster under test, and a track selection element to associate tracks provided by another sensor based on belief functions generated from the feature vectors, the track selection element to select one of the tracks corresponding to a track of interest,

wherein the feature vectors comprise one or more of a cluster count feature vector (N), a cluster population density feature vector (P), a cluster proximity feature vector (r), a cluster-weighted centroid feature vector (L) and a cluster scattering feature vector (θ

).

29. The system of claim 28 wherein the overhead sensor comprises a satellite to detect a launch of the missile and to track a plume of the missile through burnout.

30. The system of claim 28 wherein the track selection element is to generate the belief functions (μ

) from corresponding feature vectors of both sensors for each of a plurality of cluster as clusters-under-test, and

wherein the track selection element is to assign an association class based on the belief functions and a plurality of ratings for selection of a track cluster having a high-probability of including the track of interest.

31. The system of claim 30 wherein the interceptor further comprises a track state estimation element to generate a track-state vector for each of plurality of the objects detected within a threat object cloud, the objects being detected by on-board sensor circuitry of the sensor, each track-state vector comprising velocity, position and the uncertainty,

wherein the track clustering element is to cluster the tracks based associated track-state vectors having overlapping uncertainty regions, the uncertainty regions based on the covariance associated with each track-state vector,

wherein the other sensor is a ground-based tracking sensor, and wherein radar tracks of a plurality of radar-tracked objects are received from the ground based sensor,

wherein one of the radar-tracks is designated as the track of interest by the ground-based tracking sensor,

wherein the track clustering element is to further cluster the radar-tracks of the radar-tracked objects having overlapping uncertainty regions with the track of interest to generate a track-cluster of interest, and

wherein the feature generating element is to further generate feature vectors for the track-cluster of interest in two or more dimensional directions with respect to the track of interest.

32. A machine-accessible medium that provides instructions, which when accessed, cause a machine to perform operations comprising:

clustering tracks of objects to generate track clusters based on an uncertainty associated with each track;

generating feature vectors for the track clusters in one or more directions with respect to a cluster under test, the feature vectors comprising one or more of a cluster count feature vector (N), a cluster population density feature vector (P), a cluster proximity feature vector (r), a cluster-weighted centroid feature vector (L) and a cluster scattering feature vector (θ

);

associating tracks provided by another sensor based on belief functions generated from the feature vectors; and

selecting one of the tracks corresponding to a track of interest.