Target detection system using trained and untrained detection and methods therefor

• US 20040047492A1
• Filed: 06/20/2003
• Published: 03/11/2004
• Est. Priority Date: 06/20/2002
• Status: Active Grant

• Current Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Original Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Inventors: Mahalanobis, Abhijit, Muise, Robert
• Granted Patent: US 7,421,090 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes

Abstract

A target detection system and method is disclosed that uses a trained detection component and an untrained detection component that enhances image data for candidate target detection. The trained detection component separates candidate targets from clutter within the image data using correlation filters trained from an image library. The image library includes target and clutter images that are used to tune the correlation filters. The untrained detection component separates the candidate targets from the clutter by suppressing clutter using a Fourier frequency transform of the image data. Anomalies are detected in the frequency domain, and retained to highlight candidate targets.

Claims

1. A target detection system that receives image data, comprising:

a trained detection component that separates candidate target regions from clutter within the image data using a plurality of correlation filters trained from an image library;

an untrained detection component that separates the candidate target regions from the clutter by enhancing the image data using a frequency transform; and

an output highlighting the candidate target regions.

2. The target detection system of claim 1, further comprising an input device to capture the image data.

3. The target detection system of claim 1, wherein the image data comprises pixels.

4. The target detection system of claim 1, wherein the image library comprises target images and clutter images.

5. A method for enhancing imagery for target detection, comprising:

receiving image data;

determining candidate target regions from clutter within the image data using linear correlation filters, wherein the linear correlation filters are trained using an image library;

determining the candidate target regions from the clutter by suppressing the clutter in a frequency domain transform of the image data; and

identifying peak detections from an output generated by the determining steps, wherein the peak detections correspond to candidate targets.

6. The method of claim 5, wherein the determining steps are executed in a parallel manner.

7. The method of claim 5, wherein the first determining step includes identifying the candidate target regions by determining energies from correlating the linear correlation filters with the image data.

8. The method of claim 5, wherein the second determining step includes identifying anomalous components within the frequency domain transform of the image data.

9. A system for performing trained detection operations on image data, comprising:

linear correlation filters trained from an image library, wherein each of the linear correlation filters is applied to the image data to produce target and clutter filter response energies;

a summation component to determine the difference between the target filter response energies and the clutter filter response energies; and

a peak detector to detect a candidate target according to the difference between the target filter response energies and the clutter filter response energies.

10. The system of claim 9, wherein the difference is above a specified amount corresponding to the candidate target.

11. The system of claim 9, wherein the linear correlation filters are trained according to target images within the image library.

12. The system of claim 9, wherein the linear correlation filters are trained according to clutter images within the image library.

13. The system of claim 9, further comprising a transformation matrix comprising the linear correlation filters, wherein the transformation matrix is applied linearly to the image data.

14. The system of claim 9, further comprising an output map indicating the differences between the target response energies and the clutter response energies.

15. The system of claim 9, wherein the target response energies are derived from correlating pixels within the image data to target linear correlation filters.

16. The system of claim 9, wherein the clutter response energies are derived from correlating pixels within the image data to clutter linear correlation filters.

17. A method for enhancing image data for target detection, comprising:

tuning linear correlation filters with image chips from an image library, wherein the image chips include target and clutter images;

determining energy in filter responses by correlating the linear correlation filters with the image data; and

composing an output detection map of the combined filter responses to denote candidate targets and clutter.

18. The method of claim 17, further comprising detecting peaks within the output detection map for indicate the candidate targets and the clutter.

19. The method of claim 17, wherein the tuning step includes:

receiving the image chips from the image library;

calculating basis functions of the image chips;

computing detection metrics for the basis functions; and

generating the linear correlation filters according to the detection metrics.

20. The method of claim 19, further comprising selecting the basis functions according to the detection metrics.

21. The method of claim 19, further comprising separating the basis functions into target basis functions and clutter basis functions.

22. A method for enhancing a target image in a detection system, comprising:

transforming portions of said image into a frequency domain representation;

identifying anomalous components of said representation;

retaining said anomalous components; and

reverse transforming said representation.

23. The method of claim 22, further comprising receiving said image from an input device.

24. The method of claim 23, wherein the detection system includes an untrained detection component.

25. The method of claim 22, wherein said transforming includes applying a Fourier transform to a column within said image, wherein said frequency domain representation includes said transform of the column.

26. The method of claim 25, further comprising placing said transform of said column into a frequency bin.

27. The method of claim 26, further comprising normalizing said frequency bin, wherein said frequency bin includes said anomalous components.

28. The method of claim 22, further comprising suppressing a non-anomolous component.

29. The method of claim 28, wherein said suppressing includes zeroing said component.

30. A method for performing frequency anomaly detection, comprising:

receiving image data;

applying a frequency transform to a column within said image data;

placing said transformed column in a frequency bin of a sub-image;

normalizing components within said frequency bin;

determining whether said frequency bin includes anomalous components;

retaining said anomalous components; and

reverse frequency transforming said anomalous components.

31. The method of claim 30, further comprising suppressing non-anomalous components within said subimage.

32. The method of claim 30, wherein said sub-image comprises rows, said rows correlating to said frequency bins.

33. A method for detecting a target from an image, comprising:

receiving said image;

transforming said image into a frequency representation; and

selecting components of said frequency representation that meet a specified threshold, wherein said components relate to said target.

34. The method of claim 33, further comprising reverse transforming said frequency representation to identify said target.

35. The method of claim 33, wherein said selecting includes identifying components within said frequency representation having anomalies.

36. The method of claim 35, further comprising normalizing said components.

37. A system for enhancing imagery for target detection, comprising:

means for receiving image data;

means for determining candidate target regions from clutter within the image data using a set of linear correlation filters, wherein the linear correlation filters are trained using an image library;

means for determining the candidate target regions from the clutter by suppressing the clutter in a frequency domain transform of the image data; and

means for identifying peak detections from an output generated by the determining steps, wherein the peak detections correspond to candidate targets.

38. A system for enhancing a target image from image data for target detection, comprising:

an untrained detection component that transforms the image data into a frequency domain representation, and identifies anomalous components within the frequency domain representation, wherein the frequency domain representation is reversed transformed retaining the anomalous components;

an output containing portions of the image data corresponding to the anomalous components.

39. A computer program product comprising a computer useable medium having computer readable code embodied therein for enhancing imagery for target detection, the computer program product adapted when run on a computer to effect steps including:

receiving image data;

determining candidate target regions from clutter within the image data using a plurality of linear correlation filters, wherein the linear correlation filters are trained using an image library;

determining the candidate target regions from the clutter by suppressing the clutter in a frequency domain transform of the image data; and

identifying peak detections from an output generated by the determining steps, wherein the peak detections correspond to candidate targets.

40. A computer program product comprising a computer useable medium having computer readable code embodied therein for enhancing image data for target detection, the computer program product adapted when run on a computer to effect steps including:

tuning linear correlation filters with image chips from an image library, wherein the image chips include target and clutter images;

determining energy in filter responses by correlating the linear correlation filters with the image data; and

composing an output detection map of the combined filter responses to denote candidate targets and clutter.

41. A computer program product comprising a computer useable medium having computer readable code embodied therein for enhancing a target image for target detection, the computer program product adapted when run on a computer to effect steps including:

transforming portions of said image into a frequency domain representation;

identifying anomalous components of said representation;

retaining said anomalous components; and

reverse transforming said representation.