Classification of range profiles

• US 9,514,361 B2
• Filed: 04/18/2011
• Issued: 12/06/2016
• Est. Priority Date: 04/21/2010
• Status: Active Grant

• Current Assignee: BAE Systems Plc
• Original Assignee: BAE Systems Plc
• Inventors: Miller, Robert James
• Primary Examiner(s): Kim, Eunhee
• Application Number:
  US13/642,386
• Publication Number:
  US 20130041640A1
• Time in Patent Office: 2,059 Days
• Field of Search: 703/2
• US Class Current: 1/1
• CPC Class Codes: G01S 7/41 using analysis of echo signal for target characterisation; Target signature; Target cross-section
• CPC Class Codes: G06F 18/2415 based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
• CPC Class Codes: G06F 18/295 Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
• CPC Class Codes: G06F 2218/14 by matching peak patterns
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes

Abstract

A method and apparatus are provided for classifying range profiles, generated for example by a radar, lidar or sonar. In the method, each in a set of objects of interest is modeled with a probabilistic model. The probabilistic model represents the probabilities of occurrence of different possible sequences of distances between selected features of the object, in different orientations, that are likely to result in peaks of backscatter in a range profile of the object. The probabilistic model is derived from a first probabilistic representation of each selected feature, generated to represent the uncertainty in locating the feature and the uncertainty in observing the feature in a range profile. Classification is achieved by calculating, for each probabilistic model, the probability that the model would generate a given sequence of distances between observed backscatter events in a given range profile. The model generating the given sequence with the greatest probability identifies the object likely to have produced the given range profile. Preferably, the probabilistic models comprise Hidden Markov Models (HMMs).

Claims

1. A method for classifying range profiles, comprising:

gathering non-sensor based sources of information giving structural details of objects of interest;

selecting, from the non-sensor based sources of information, features of the objects of interest that are configured to appear most prominent as peaks of backscatter in a sensor based observation of the objects of interest;

generating, for each of the objects of interest, a probabilistic model representing, for one or more different orientations of the respective object of interest, possible sequences of distances between the features of the respective object of interest selected from the non-sensor based sources of information that are configured to appear as distinct peaks of backscatter in sensor based range data for the object, wherein the possible sequences of distances are derived from a first probabilistic representation of each of the features of the respective object of interest;

classifying a given sensor based range profile by deriving an observed sequence of distances from the spacing of distinct peaks of backscatter in the given sensor based range profile and by calculating, for each of the probabilistic models, a probability that the respective probabilistic model represents the observed sequence of distances, wherein the object of interest represented by the probabilistic model that represents the observed sequence of distances with the greatest probability is associated with the given sensor based range profile; and

generating classification results for at least one of the probabilistic models so as to predict the potential performance of a classifier.

2. The method of claim 1 wherein generating the probabilistic model comprises:

generating a first probabilistic representation defining an uncertainty in locating each of the features of the respective object of interest selected from the non-sensor based sources of information and an uncertainty in whether or not each said feature is observed;

deriving, for one or more orientations of the object, inferred sequences of said features which are observable as distinct peaks of backscatter in sensor based range data for the object, and the possible sequences of distances between said features in each inferred sequence; and

generating, from the first probabilistic representation and the possible sequences of distances, a probabilistic model for the object from which the probability is calculated that an observed sequence of distances corresponds to a possible sequence derived from the first probabilistic representation.

3. The method according to claim 1, wherein the probabilistic model for the object comprises a set of one or more Hidden Markov Models (HMM), each HMM defining, for a different orientation of the object, probabilities for the possible sequences of distances between the features of the respective object of interest selected from the non-sensor based sources of information.

4. The method according to claim 3, wherein generating the probabilistic model further comprises estimating the orientation of an object represented in the given sensor based range profile, and wherein said one or more probabilistic models comprise HMMs generated for objects having said estimated orientation.

5. The method according to claim 3, further comprising:

generating a plurality of examples of a range profile for one or more of said objects of interest from their respective HMMs and using a simulation technique to generate the classification results for a particular set of HMMs so as to predict the potential performance of the classifier.

6. The method according to claim 1, wherein said sensor based range data and the given sensor based range profile relate to and are generated by a high range resolution radar, and wherein said distinct peaks correspond to peaks in backscattered radiation from the object.

7. The method according to claim 1, wherein said sensor based range data and the given sensor based range profile relate to and are generated by at least one of a lidar and a sonar.

8. The method according to claim 1, wherein the non-sensor based sources of information include at least one of:

engineering drawings, photographs of the objects of interest, and scale models of the objects of interest.

9. The method according to claim 1, further comprising determining locations of observable features as linear distances from a reference point to the respective observable features, wherein the distances between observable features of the object are based at least in part of the locations of the observable features.

10. The method according to claim 9, further comprising determining an angular field of visibility of each selected feature.

11. The method according to claim 9, wherein the observed sequence of distances includes a set of consecutive differences in the locations of observable features along a length of the object.

12. The method according to claim 9, wherein the observed sequence of distances includes a set of differences in the locations of observable features with respect to the reference point.

13. A non-transitory computer program product encoded with instructions that, when executed by one or more processors, causes a process to be carried out, the process comprising:

gathering non-sensor based sources of information giving structural details of objects of interest;

selecting, from the non-sensor based sources of information, features of the objects of interest that are configured to appear most prominent as peaks of backscatter in a sensor based observation of the objects of interest;

generating, for each of the objects of interest, a probabilistic model representing, for one or more different orientations of the respective object of interest, possible sequences of distances between the features of the respective object of interest selected from the non-sensor based sources of information that are configured to appear as distinct peaks of backscatter in sensor based range data for the object, wherein the possible sequences of distances are derived from a first probabilistic representation of each of the features of the respective object of interest;

classifying a given sensor based range profile by deriving an observed sequence of distances from the spacing of distinct peaks of backscatter in the given sensor based range profile and by calculating, for each of the probabilistic models, a probability that the respective probabilistic model represents the observed sequence of distances, wherein the object of interest represented by the probabilistic model that represents the observed sequence of distances with the greatest probability is associated with the given sensor based range profile; and

generating classification results for at least one of the probabilistic models so as to predict the potential performance of a classifier.

14. The non-transitory computer program product according to claim 13, wherein the process further comprises:

generating a first probabilistic representation defining an uncertainty in locating each of the features of the respective object of interest selected from the non-sensor based sources of information and an uncertainty in whether or not each said feature is observed;

deriving, for one or more orientations of the object, inferred sequences of said features which are observable as distinct peaks of backscatter in sensor based range data for the object, and the possible sequences of distances between said features in each inferred sequence; and

generating, from the first probabilistic representation and the possible sequences of distances, a probabilistic model for the object from which the probability is calculated that an observed sequence of distances corresponds to a possible sequence derived from the first probabilistic representation.

15. The non-transitory computer program product according to claim 13, wherein the probabilistic model for the object comprises a set of one or more Hidden Markov Models (HMM), each HMM defining, for a different orientation of the object, probabilities for the possible sequences of distances between the features of the respective object of interest selected from the non-sensor based sources of information.

16. The non-transitory computer program product according to claim 15, wherein generating the probabilistic model further comprises estimating the orientation of an object represented in the given sensor based range profile, and wherein said one or more probabilistic models comprise HMMs generated for objects having said estimated orientation.

17. The non-transitory computer program product according to claim 15, wherein the process further comprises:

generating a plurality of examples of a range profile for one or more of said objects of interest from their respective HMMs and using a simulation technique to generate the classification results for a particular set of HMMs so as to predict the potential performance of the classifier.

18. The non-transitory computer program product according to claim 13, wherein said sensor based range data and the given sensor based range profile relate to and are generated by a high range resolution radar, and wherein said distinct peaks correspond to peaks in backscattered radiation from the object.

19. The non-transitory computer program product according to claim 13, wherein said sensor based range data and the given sensor based range profile relate to and are generated by at least one of a lidar and a sonar.

20. The non-transitory computer program product according to claim 13, wherein the non-sensor based sources of information include at least one of:

engineering drawings, photographs of the objects of interest, and scale models of the objects of interest.