IDENTITY TRACKING PROCESS AND SYSTEM

• US 20110098994A1
• Filed: 05/05/2009
• Published: 04/28/2011
• Est. Priority Date: 05/06/2008
• Status: Active Grant

• Current Assignee: Leonardo S.p.A.
• Original Assignee: Selex Galileo Limited (Leonardo S.p.A.)
• Inventors: Cade, Neil
• Granted Patent: US 8,638,642 B2
• Time in Patent Office: Days
• US Class Current: 703/2
• CPC Class Codes: G01V 1/001 Acoustic presence detection

Abstract

A method of identifying and tracking a target is described, in which seismic data relating to a target is passively detected and processed using statistical means. The statistical manipulation of the data includes frequency information extraction, dynamical mixture model construction based on existing known data and identification of an unknown target by the convergence of this model to a state characteristic of that target.

Claims

1. A process for identifying and tracking a target, comprising:

detecting emissions emitted by the target;

processing emissions data detected using a series of statistical manipulation; and

identifying and tracking the target in which the emissions detected are detected passively by at least one sensor.

2. A process as recited in claim 1, in which the emissions detected are acoustic in nature and are detected at some distance from the target after passage of acoustic vibrations though intervening solid liquid or gaseous medium.

3. A process as recited in claim 2, in which the emissions detected are of seismic nature and are detected passively by seismic sensors.

4. A process as recited in claim 1, in which the emissions detected are of electromagnetic nature and are detected passively by broad band electromagnetic sensors.

5. A process as recited in claim 1, in which the statistical manipulations comprise:

filtering the seismic data detected;

extracting statistical frequency information;

constructing a statistical mixture model describing the emission characteristics of the target observed in a range of different environments; and

constructing a dynamic mixture model construction based on existing known data and identification of an unknown target by convergence of the model to a state characteristic of the target.

6. A process as described in claim 5, where the the dynamic model is constructed with two parts:

a first part comprising separate models for individual targets, and a second part comprising a dynamic model incorporating stochastic switching between models of all targets of interest.

7. A process as described recited in claim 5, in which constructions of the statistical mixture model and the dynamic mixture model are performed off-line on data collected from known examples of targets under a range of representative conditions, and where the identifying is performed on-line using previously constructed models and data collected under circumstances where the target is unknown.

8. A process as recited in claim 1, in which the dynamic model is a Hidden Markov model where hidden states correspond to individual target models and transitions between the states correspond to changes of target identity, with actual identity being obtained by comparisons of probabilities of the identity states of the model.

9. A process as recited in claim 8, in which the states of the Hidden Markov model are associated with individual model components, and the transitions involve both changes of identity and changes of component so that an identification is in terms of both the target identity and the state of the target, constituting both identification and generalized tracking.

10. A process as recited in claim 1, in which statistical frequency information is a cumulative probability density or discrete cumulative probability distribution.

11. A process as recited in claim 10, in which statistical frequency information is obtained by averaging over occurrences of frequencies obtained by counting zero-crossings.

12. A process as recited in claim 8, in which individual model components in the Hidden Markov model are associated with motion of a vehicle at different distances from the sensor or on different terrains so that identification is in terms of both the target identity and location of the target.

13. A process as recited in the claim 12, in which the location and identity information from an array of sensors can be used to track the detected vehicle in space, possibly using map data to provide terrain information to remove any ambiguities.

15. A process as recited in claim 2, in which the statistical manipulations comprise:

filtering the seismic data detected;

extracting statistical frequency information;

constructing a statistical mixture model describing the emission characteristics of the target observed in a range of different environments; and

constructing a dynamic mixture model construction based on existing known data and identification of an unknown target by convergence of the model to a state characteristic of the target.

16. A process as recited in claim 3, in which the statistical manipulations comprise:

filtering the seismic data detected;

extracting statistical frequency information;

constructing a statistical mixture model describing the emission characteristics of the target observed in a range of different environments; and

constructing a dynamic mixture model construction based on existing known data and identification of an unknown target by convergence of the model to a state characteristic of the target.

17. A process as recited in claim 4, in which the statistical manipulations comprise:

filtering the seismic data detected;

extracting statistical frequency information;

constructing a statistical mixture model describing the emission characteristics of the target observed in a range of different environments; and

constructing a dynamic mixture model construction based on existing known data and identification of an unknown target by convergence of the model to a state characteristic of the target.

18. A process as recited in claim 6, in which constructions of the statistical mixture model and the dynamic mixture model are performed off-line on data collected from known examples of targets under a range of representative conditions, and where the identifying is performed on-line using previously constructed models and data collected under circumstances where the target is unknown.

19. A process as recited in claim 2, in which the dynamic model is a Hidden Markov model where hidden states correspond to individual target models and transitions between the states correspond to changes of target identity, with actual identity being obtained by comparisons of probabilities of the identity states of the model.

20. A process as recited in claim 3, in which the dynamic model is a Hidden Markov model where hidden states correspond to individual target models and transitions between the states correspond to changes of target identity, with actual identity being obtained by comparisons of probabilities of the identity states of the model.

21. A process as recited in claim 4, in which the dynamic model is a Hidden Markov model where hidden states correspond to individual target models and transitions between the states correspond to changes of target identity, with actual identity being obtained by comparisons of probabilities of the identity states of the model.

14. (canceled)