Object detection system with minimum-spanning gradient filter for scene clutter suppression

• US 5,805,742 A
• Filed: 08/16/1995
• Issued: 09/08/1998
• Est. Priority Date: 08/16/1995
• Status: Expired due to Term

• Current Assignee: Northrop Grumman Systems Corporation (Northrop Grumman Corporation)
• Original Assignee: TRW Limited (Zeppelin - Stiftung der Stadt Friedrichshafen)
• Inventors: Whitsitt, Stephen J.
• Primary Examiner(s): Couso, Yon
• Assistant Examiner(s): Patel, Jayanti K.
• Application Number:
  US08/624,304
• Time in Patent Office: 1,119 Days
• Field of Search: 382/103, 382/278, 382/275, 382/260, 342/352, 348/169, 250/203.6, 250/332, 250/342, 250/339.14
• US Class Current: 382/275
• CPC Class Codes: G01S 3/781 Details
• CPC Class Codes: G01S 3/786 the desired condition being maintained automatically
• CPC Class Codes: G02B 5/205 Neutral density filters

Abstract

A minimum-spanning gradient filter used to suppress clutter in a target detection and tracking system. The minimum-spanning gradient filter uses subspace projection clutter suppression techniques, but does not require eigenanalysis. Model frames of data from a sensor array are stored in a memory in which the model frames do not include targets that are being detected. At start-up, a Gram-Schmidt system generates a series of unit vectors that define a clutter subspace where most of the clutter in the model frames will reside. Current frames of data from the sensor array and unit vectors from the Gram-Schmidt system are applied to a subspace projection system that removes the clutter subspace from the current frames of data. Once the original clutter subspace has been generated, a plane smoothing system updates the clutter subspace as new frames of data are added to the model frames. Current frames of data sent to the subspace projection system are centered. Additionally, a standard deviation for the pixel intensities in each frame of data is applied to a standard deviation threshold to prevent pixels that fall below the threshold from being updated in the clutter subspace. A glint threshold detector is provided to eliminate pixels above a predetermined glint threshold so that they do not contribute to the updated clutter subspace.

Claims

1. A clutter suppression system for suppressing clutter in an object detection system, said object detection system sensing a scene to determine if an object of interest is present, said clutter suppression system comprising:

frame generation means for generating a series of frames of data where the frames of data represent images of the scene in which clutter is to be suppressed;

model storage means for storing model frames of the scene based on the series of frames of data, wherein the model frames of the scene are constructed when the object of interest does not exist in the scene;

clutter subspace means for generating a clutter subspace of the scene at a predetermined time from the model frames that defines substantially all of the clutter in the scene; and

subspace projection means for generating frames of data of the scene in which substantially all of the clutter has been removed, said subspace projection means being responsive to the series of frames of data and the clutter subspace from the clutter subspace means, said subspace projection means measuring the clutter subspace and removing the clutter from the frames of data to generate substantially clutter free frames of data to be detected for the presence of the object, said subspace projection means being capable of removing clutter from the frames of data both prior to and later than the predetermined time the clutter subspace is generated by the clutter subspace means.

2. The clutter suppression system according to claim 1 further comprising centering means for centering the model frames of data, said centering means including pixel intensity averaging means for averaging the intensities of the pixels in the frames of data, said subspace projection means being responsive to centered pixel intensities of the model frames of data.

3. The clutter suppression system according to claim 1 wherein the clutter subspace means includes a Gram-Schmidt processing system that determines a set of clutter subspace unit vectors by a Gram-Schmidt procedure, wherein the unit vectors define the clutter subspace in the model frames.

4. The clutter suppression system according to claim 3 wherein the Gram-Schmidt processing system is responsive to centered pixel intensities from a centering means for centering pixel intensities of pixels in the model frames of data.

5. The clutter suppression system according to claim 3 wherein the Gram-Schmidt processing system determines the clutter subspace unit vectors by the relationship:

##EQU17## where, S_i is a centered scene vector for time i in the model frames;

C is a unit vector for the clutter subspace;

i is an index corresponding to a time sample;

j is a summation index; and

T is a transpose operation.

6. The clutter suppression system according to claim 1 wherein the clutter subspace means includes a plane smoothing system that determines a set of clutter subspace unit vectors, said plane smoothing being responsive to the frames of data from the subspace projection means and the frames of data from the frame generation means so as to generate updated clutter subspace unit vectors in order to update the clutter subspace.

7. The clutter suppression system according to claim 6 wherein the plane smoothing system smoothes the rotation of the clutter subspace relative to an original clutter subspace as frames of data are added to the model frames by using a recursive filter approach.

8. The clutter suppression system according to claim 1 further comprising clutter sniffer means for determining the standard deviation of pixels in the model frames of data, said clutter sniffer means being responsive to pixel intensities in the model frames so as to generate the standard deviation of the model frames, wherein the standard deviation of each pixel is applied to a pixel threshold so that if the standard deviation of the pixel is below the pixel threshold the pixel is removed from the clutter subspace.

9. The clutter suppression system according to claim 1 further comprising glint threshold means for determining if pixels in the frames of data have an absolute intensity above a predetermined glint threshold, wherein if the pixel intensity is above the glint threshold the pixel is removed from the clutter subspace.

10. The clutter suppression system according to claim 1 further comprising object detection storage means and delay storage means, said object detection storage means storing current frames of data to be tested for the present of an object, said delay storage means storing past frames of data that include a detected object of interest, wherein the model storage means stores model frames of data that do not include the object.

11. The clutter suppression system according to claim 1 wherein the subspace projection means generates frames of data substantially free of clutter by the relationship:

##EQU18## where, S is a centered scene vector in the model frames;

C is unit vector for the clutter subspace;

k is a time sample;

i is an index corresponding to a time sample; and

T is a transpose operation.

12. A clutter suppression system for suppressing clutter in an object detection system, said object detection system sensing a scene to determine if an object of interest is present, said clutter suppression system comprising:

a model storage device that stores model frames of the scene, wherein the model frames of the scene are constructed when the object of interest does not exist in the scene;

a Gram-Schmidt processing system being responsive to the model frames of the scene, said Gram-Schmidt processing system generating a set of clutter subspace unit vectors directly from the model frames that define a clutter subspace in the model frames where substantially all of the clutter in the model frames resides; and

a subspace projection system that is responsive to frames of data of the scene and the clutter subspace unit vectors from the Gram-Schmidt processing system, said subspace projection system measuring the clutter subspace and then removing the clutter subspace from the frames of data to remove the clutter in the current frames of data.

13. The clutter suppression system according to claim 12 wherein the Gram-Schmidt processing system determines the clutter subspace unit vectors by the relationship:

##EQU19## where, S_i is a centered scene vector for time i in the model frames;

C is a unit vector for the clutter subspace;

i is an index corresponding to a time sample;

j is a summation index; and

T is a transpose operation.

14. A clutter suppression system for suppressing clutter in an object detection system, said clutter detection system sensing a scene to determine if an object of interest is present, said clutter suppression system comprising:

a model storage device that stores model frames of data of the scene, wherein the model frames of the scene are constructed when the object of interest does not exist in the scene;

a plane smoothing system responsive to the model frames of data from the model storage device, said plane smoothing system determining a set of clutter subspace unit vectors that define a clutter subspace in the model frames, said plane smoothing system updating the clutter subspace by rotating the clutter subspace in a direction of a new model frame, said plane smoothing system controlling the amount of rotation of the clutter subspace to reduce the effects of random noise; and

a subspace projection system responsive to the model frames of data and the clutter subspace unit vectors, said subspace projection system measuring the clutter subspace and then removing the clutter subspace from the frames of data to remove the clutter in the current frames of data.

15. The clutter suppression system according to claim 14 further comprising a Gram-Schmidt processing system that determines an original set of clutter subspace unit vectors by a Gram-Schmidt procedure.

16. The clutter suppression system according to claim 14 wherein the plane smoothing system controls the amount of rotation of the clutter subspace by allowing the rotation to be in proportion to the amplitude of the difference between the new model frame and its projection onto the clutter subspace.

17. The clutter suppression system according to claim 14 wherein the plane smoothing system controls the amount of rotation of the clutter subspace by allowing the rotation to be a fraction of an angle between the new model frame and the clutter subspace.

18. A method of suppressing clutter in an object detection system, said object detection system sensing a scene to determine if an object of interest is present, said method comprising the steps of:

providing a series of frames of data representing images of the scene in which clutter is to be suppressed;

using the frames of data to generate a series of model frames of the scene, wherein the model frames of the scene are constructed when the object of interest does not exist in the scene;

using the model frames of the scene to generate a clutter subspace at a predetermined time that defines substantially all of the clutter in the scene; and

measuring the clutter subspace and removing the clutter subspace from the frames of data so as to remove substantially all of the clutter in the frames of data, said step of removing the clutter subspace including the capability of removing clutter from the frames of data both prior to and later than the predetermined time the clutter subspace is generated.

19. The method according to claim 18 further comprising the step of centering the model frames of data so as to provide centered pixel intensities of the model frames of data to be used by the subspace projection technique.

20. The method according to claim 18 wherein the step of generating a clutter subspace includes using a Gram-Schmidt process to determine a set of clutter subspace unit vectors that define the clutter subspace.

21. The method according to claim 18 wherein the step of generating a clutter subspace includes using a plane smoothing technique that determines a set of clutter subspace unit vectors that define the clutter subspace, said step of using a plane smoothing technique including smoothing the rotation of the clutter subspace as new frames are added to the model frames of data.

22. The method according to claim 18 further comprising the step of determining the standard deviation of the pixels in the model frames of data and applying the standard deviation to a clutter threshold to determine if the standard deviation of the pixels is below the clutter threshold, wherein if the standard deviation is below the clutter threshold the pixel is removed from the clutter subspace.

23. The method according to claim 18 further comprising the steps of determining if the pixels in the frames of data have an absolute intensity above a predetermined glint threshold, wherein if the pixel intensity is above the glint threshold the pixel is removed from the clutter subspace.

24. The method according to claim 18 further comprising the steps of testing current frames of data for a target and delaying the frames of data that are to be added to the model frames if a target is detected in the current frames of data.

25. A clutter suppression system for suppressing clutter in an object detection system, said object detection system sensing a scene to determine if an object of interest is present, said clutter suppression system comprising:

a system that generates a series of frames of data that represent images of the scene in which clutter is to be suppressed, each frame of data including a fixed number of pixels that define the image;

a storage device that stores model frames of the scene based on the series of frames of data where the model frames of the scene are constructed when the object of interest does not exist in the scene;

a system that generates a clutter subspace of the scene from the model frames of the scene where the clutter subspace defines substantially all of the clutter in the scene; and

a subspace projection system that generates frames of data of the scene in which substantially all of the clutter has been removed, the subspace projection system being responsive to the series of frames of data and the clutter subspace from the clutter subspace system, said subspace projection system modelling the clutter subspace and removing the clutter from the frames of data to generate substantially clutter free frames of data to be detected for the presence of the object, wherein the clutter subspace is modelled as a set of gradient vectors for each pixel, said gradient vectors being relatively constant over a sequence of frames of data used to generate the model frames.