Face identification verification using 3 dimensional modeling

• US 7,421,097 B2
• Filed: 05/27/2003
• Issued: 09/02/2008
• Est. Priority Date: 05/27/2003
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Hamza, Rida M., Bazakos, Michael E., Cooper, Murray J.
• Primary Examiner(s): KIM, CHONG R
• Application Number:
  US10/446,521
• Publication Number:
  US 20040240711A1
• Time in Patent Office: 1,925 Days
• Field of Search: None
• US Class Current: 382/118
• CPC Class Codes: G06V 20/64 Three-dimensional objects
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A checkpoint screening system uses multiple cameras to provide images at different angles of a person to be screened. Different features extracted from the images at different angles are compared to a profile associated with the person to be screened. In one embodiment, the person first provides an ID, such as a drivers license or other identification, and the profile is retrieved. If a match is detected, the person may continue through the checkpoint. If no match is detected, the person may be directed to a different path through the checkpoint for further verification of identity. A registration process is used to enroll members and obtain a member profile. Three camera angles are utilized to provide a three dimensional image. Each image is independently compared against the profile and the decisions from such comparisons are weighted. The profile is regressively updated and weighted if a match is detected.

Claims

1. A method of matching faces, the method comprising:

obtaining a front view image of a face to be matched;

obtaining a profile view image of the face to be matched from an angle off the front view;

generating image features from each of the view images wherein front view image and the profile view image from an angle off the front view comprise different features;

combining the image features from each of the view images into a combined feature set;

comparing the combined feature set to a profile, wherein features from the profile view image are used to normalize the features obtained from the front view image prior to comparing the combined feature set to the profile; and

determining a match based on the compared image features.

2. The method of claim 1 wherein the profile is updated if a match is determined.

3. The method of claim 2 wherein the profile comprises a set of features, and wherein such features are updated based on features identified in the view images.

4. The method of claim 3 wherein a feature comprises multiple weighted data points.

5. The method of claim 4 wherein most recent data points are more heavily weighted.

6. The method of claim 1 wherein one feature of the image from an angle comprises distance measurements of nose, ear and other landmarks to a relative nodal point in the face landscape.

7. The method of claim 1 and further comprising obtaining a further view image of the face to be matched from an angle off the front view in the other direction from the other angled view.

8. The method of claim 1, wherein at least one feature from the profile view image is not available from a valid front view image.

9. The method of claim 8 wherein a valid front view image may be obtained from a camera positioned within 20 degrees of straight on to the face.

10. The method of claim 1 wherein normalizing the features comprises a perspective transformation to provide a desired orientation in space of a frontal plane of the face.

11. A method of matching faces, the method comprising:

obtaining multiple front view images of a face to be matched;

obtaining a profile view image of the face to be matched from an angle off the front view, wherein at least one feature from the profile view image is not available from a valid front view image;

selecting one of the multiple front view images;

generating image features from each of the view images wherein the front view image and the image from an angle off the front view comprise different features;

combining the generated image features from each of the view images into a combined feature set;

comparing the combined feature set to a profile, wherein features from the profile view image are used to normalize the features obtained from the selected front view image prior to comparing the combined feature set to the profile; and

determining a match based on the compared image features.

12. A system for matching faces, the system comprising:

means for obtaining a front view image of a face to be matched;

means for obtaining a profile view image of the face to be matched from an angle off the front view;

means for generating different image features from each of the view images;

means for normalizing the front view image features as a function of the profile view image features and comparing the normalized image features to a profile; and

means for determining a match based on the compared image features.

13. A method of checking identity at a security checkpoint, the method comprising:

reading an ID for a person desiring to proceed through the checkpoint;

obtaining multiple laterally spaced views of the person, wherein the laterally spaced views comprise a front view and at least one profile view;

deriving feature sets from each of the views, wherein at least some of the features are not obtainable from any frontal view of the person;

combining the derived feature sets into a single feature set containing different features from each of the views;

retrieving pre-existing feature sets corresponding to information on the ID; and

comparing the derived feature set to the pre-existing feature sets to determine whether they match, wherein features from at least one laterally spaced view are used to normalize the features obtained from at least one front view image prior to comparing the single feature set to the pre-existing feature sets.

14. The method of claim 13 and further comprising directing the person to a manual security station if there is no match.

15. The method of claim 13 and further comprising enrolling a person to create the pre-existing feature set.

16. The method of claim 15 and further comprising determining whether a person is an expired member.

17. The method of claim 13 and further comprising updating the pre-existing feature set upon detecting a match.

18. The method of claim 17 and further comprising temporally weighting feature sets.

19. The method of claim 13 wherein at least one of the views is a profile view.

20. The method of claim 13 wherein the laterally spaced views comprise a frontal view and two profile views on opposite sides from the frontal view.

21. The method of claim 20 wherein the profile views are spaced between approximately 20 and 90 degrees from the frontal view.

22. The method of claim 13 and further comprising detecting motion between features.

23. A security checkpoint system comprising:

means for reading an ID for a person desiring to proceed through the checkpoint;

multiple laterally spaced sensors for obtaining multiple laterally spaced views of the person, wherein the laterally spaced views comprise a front view and at least one profile view; and

a verification engine that derives a combined feature set from feature sets from each of the views wherein the combined feature set includes only different features from multiple views, normalizes front view image features of the combined feature set as a function of the one or more features from at least one laterally spaced view, retrieves pre-existing feature sets corresponding to information on the ID and compares the combined feature set to the pre-existing feature sets to determine whether they match.

24. A method of matching faces, the method comprising:

obtaining a front view image of a face to be matched;

obtaining a profile view image of the face to be matched from an angle off the front view;

generating image features from each of the view images wherein front view image and the profile view image from an angle off the front view comprise different features;

combining the image features from each of the view images into a combined feature set;

comparing the combined feature set to a profile;

determining a match based on the compared image features;

if a profile match is determined, updating such profile, wherein the profile comprises a set of features, wherein such features are updated based on features identified in the view images, wherein a feature comprises multiple weighted data points, and wherein most recent data points are more heavily weighted.