System and method for detecting and tracking features in images

• US 9,014,465 B2
• Filed: 02/21/2012
• Issued: 04/21/2015
• Est. Priority Date: 12/12/2006
• Status: Active Grant

• Current Assignee: Rutgers University
• Original Assignee: Rutgers University
• Inventors: Metaxas, Dimitris, Kanaujia, Atul
• Primary Examiner(s): CHANG, JON CARLTON
• Application Number:
  US13/401,153
• Publication Number:
  US 20130022263A1
• Time in Patent Office: 1,155 Days
• Field of Search: None
• US Class Current: 382/159
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 7/251 involving models
• CPC Class Codes: G06V 10/7553 based on shape, e.g. active shape models [ASM]
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 10/774 Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 40/171 Local features and components; Facial parts eye characteristics G06V40/18; Occluding parts, e.g. glasses; Geometrical relationships
• CPC Class Codes: G06V 40/176 Dynamic expression
• CPC Class Codes: G06V 40/20 Movements or behaviour, e.g. gesture recognition recognition of facial expressions G06V40/16

Abstract

A system and method for tracking features is provided which allows for the tracking of features that move in a series of images. A training set of images is processed to produce clustered shape subspaces corresponding to the set of images, such that non-linear shape manifolds in the images are represented as piecewise, overlapping linear surfaces that are clustered according to similarities in perspectives. A landmark-based training algorithm (e.g., ASM) is applied to the clustered shape subspaces to train a model of the clustered shape subspaces and to create training data. A subsequent image is processed using the training data to identify features in the target image by creating an initial shape, superimposing the initial shape on the target image, and then iteratively deforming the shape in accordance with the model until a final shape is produced corresponding to a feature in the target image.

Claims

1. A method for detecting blinking of a subject and tracking eye movement in a series of images, comprising the steps of:

detecting a feature in a set of images corresponding to a subject'"'"'s eye;

comparing the feature to a template of a closed eye;

determining whether the detected feature matches the template; and

if the detected feature matches the template, indicating that a subject blinked; and

training a shape model for tracking the eye of the subject using clustered shape subspaces.

2. The method of claim 1, further comprising learning the template for the eye by analyzing a plurality of initial frames of a video of a subject.

3. The method of claim 1, wherein the step of training the shape model further comprises aligning two-dimensional planar shapes in a global tangent space to cluster the shapes and learn linear subspaces for each of the clusters.

4. The method of claim 3, wherein the step of training the shape model for the eye further comprises aligning the two-dimensional planar shapes using Procrustes analysis to produce a plurality of aligned shapes.

5. The method of claim 4, wherein the step of training the shape model for the eye further comprises clustering the plurality of aligned shapes into a plurality of clusters using an Expectation-Maximization algorithm.

6. The method of claim 5, wherein the step of training the shape model for the eye further comprises generating linear subspaces from the two-dimensional planar shapes.

7. The method of claim 6, wherein the step of training the shape model for the eye further comprises learning locally linear point distribution models using Principal Component Analysis within each of the clustered shape subspaces.

8. The method of claim 7, wherein the step of training the shape model for the eye further comprises learning intensity profile models for each landmark for each of the clustered shape subspaces.

9. The method of claim 8, wherein the step of training the shape model for the eye further comprises deriving multivariate Gaussian mixture density distributions from the intensity profile models.

10. The method of claim 1, further comprising deforming an initial shape using the shape model for the eye.

11. The method of claim 10, wherein the step of deforming the initial shape further comprises constraining the deformation of the shape to match at least one of the clustered shape subspaces to produce an intermediate shape.

12. The method of claim 11, further comprising assigning a new cluster shape space to the intermediate shape by finding a nearest cluster based on a mahalanobis distance.

13. The method of claim 11, further comprising projecting the intermediate shape onto a local tangent space at a cluster mean shape.

14. The method of claim 1, further comprising tracking the feature across a plurality of subsequent images wherein the feature moves relative to frames of the subsequent images.

15. The method of claim 14, wherein the step of tracking the feature further comprises tracking the feature across the clustered shape subspaces in consecutive frames using a sum of squared intensity difference (SSID) tracker.

16. The method of claim 14, wherein the step of tracking the feature further comprises re-initializing points which have lost track by searching along a normal and maximizing profile mixture densities.

17. A method for detecting blinking of a subject and estimating a pose angle of the subject in a series of images, comprising the steps of:

detecting a feature in a set of images corresponding to a subject'"'"'s eye;

comparing the feature to a template of a closed eye;

determining whether the detected feature matches the template; and

if the detected feature matches the template, indicating that a subject blinked; and

estimating a pose angle of a subject in a target image by learning Mixture of Experts models for shapes and using feature shapes to predict tilt, pitch and yaw of a head.

18. The method of claim 17, further comprising learning the template for the eye by analyzing a plurality of initial frames of a video of a subject.

19. The method of claim 17, further comprising tracking the feature across a plurality of subsequent images wherein the feature moves relative to frames of the subsequent images.

20. The method of claim 19, wherein the step of tracking the feature further comprises re-initializing points which have lost track by searching along a normal and maximizing profile mixture densities.