Dynamic face identification

Patent metadata

Field	Value
Publication number	US-9858679-B2
Application number	US-201414532214-A
Country	US
Kind code	B2
Filing date	Nov 4, 2014
Priority date	Nov 4, 2014
Publication date	Jan 2, 2018
Grant date	Jan 2, 2018

Abstract

Systems and methods associated with dynamic face identification are disclosed. One example method includes matching a query face against a set of clusters in a dynamic collection. Matching the query face against the set of clusters may facilitate identifying a person associated with the query face. The example method also includes matching the query face against a set of images in a static gallery to identify the person. Matching the query face against the static gallery may be performed when matching the query face against the set of clusters fails to identify the person. The example method also includes updating the set of clusters in the dynamic collection using the query face.

First claim

What is claimed is:

1. A non-transitory computer-readable medium storing computer-executable instructions that when executed by a computer cause the computer to:
match a query face against a set of clusters in a dynamic collection to identify a person associated with the query face, wherein the dynamic collection includes a set of tracking targets and current features of the person in a series of facial positions and facing directions than a more limited set of images of the person in a static gallery taken at a prior point in time, and wherein images within a cluster share an attribute for identification with the query face of the person using the set of tracking targets;
match the query face against the set of images in the static gallery to identify the person, when matching the query face against the set of clusters fails to identify the person; and
update the set of clusters in the dynamic collection using the query face by adding the query face to a new member of a cluster when a distance between the query face and members of the cluster exceed a distance threshold and recalculating the attribute of the cluster.

2. The non-transitory computer-readable medium of claim 1 , where the instructions further cause the computer to:
capture a still image of the person from a frame of a video feed; and
update the set of clusters in the dynamic collection using the still image.

3. The non-transitory computer-readable medium of claim 1 , wherein the query face is a still image of the person obtained from a frame of a video feed, and when the person has not been captured by the video feed for a predetermined period of time, remove an entire set of clusters and tracking targets associated with the person from the dynamic collection.

4. The non-transitory computer-readable medium of claim 1 , where the instructions that cause the computer to update the set of clusters include instructions that cause the computer to:
modify a member of the set of clusters when a distance between the query face and the member of the set of clusters falls below a threshold; and
generate a new member of the set of clusters when distances between the query face and members of the set of clusters exceed the threshold.

5. The non-transitory computer-readable medium of claim 4 , where the distance between the query face and the member of the set of clusters is generated by determining a distance between feature vectors associated with the query face and a model face associated with the member of the set of clusters.

6. The non-transitory computer-readable medium of claim 5 , where the model face is selected from the member of the set of clusters by:
computing a centroid of feature vectors of faces of the member of the set of clusters; and
selecting a face closest to the centroid to be the model face.

7. The non-transitory computer-readable medium of claim 1 , where the dynamic collection is stored in a local memory, and where the static gallery is stored at one of a local storage media and a remote storage media.

8. The non-transitory computer-readable medium of claim 1 , where a member of the set of clusters comprises a set of images of the person that share an attribute.

9. The non-transitory computer readable medium of claim 2 , wherein by using the set of tracking targets to track the person during their movements within the video feed, the still image of the person is associated with an appropriate cluster in the dynamic collection without performing facial recognition analysis of the still image with the static gallery.

10. A system, comprising:
a data store to store a set of tracking targets and a set of image clusters in a dynamic collection, wherein each image cluster is associated with a member of the set of tracking targets and the set of image clusters in the dynamic collection includes current features of a person in a series of facial positions and facing directions than a more limited set of images of the person in a fixed image gallery taken at a prior point in time; and
a processor coupled to the data store, the processor configured to execute at least partially:
a face detection logic to update the set of tracking targets by identifying faces within frames from a video feed, and wherein images within an image cluster share an attribute for identification with the faces within frames from the video feed using the set of tracking targets;
a tracking logic to update the set of tracking targets by identifying locations of members of the set of tracking targets within the frames;
a dynamic face modeling logic to collect an image of a captured member of the set of tracking targets from the frames and to update the image cluster associated with the captured member of the set of tracking targets by adding the image of the captured member to a new member of the image cluster when a distance between the image of the captured member and members of the cluster exceed a distance threshold and recalculating the attribute of the image cluster; and
a face recognition logic to identify, using one or more of the set of image clusters and the fixed image gallery, a member of the set of tracking targets as the person associated with an image in the fixed image gallery.

11. The system of claim 10 , further comprising a frame acquisition logic to obtain the frames from the video feed and to prepare the frames for processing.

12. The system of claim 10 , where the data store is a local memory, and where the fixed image gallery is stored at one of a local data store embodied on a local storage media and a remote data store embodied on a remote storage media.

13. The system of claim 10 , comprising a purging logic to evict image clusters associated with a purged member of the set of tracking targets from the data store when the purged member of the set of tracking targets has been absent from frames for a predetermined period of time.

14. The system of claim 10 , where the face detection logic operates on every m th frame, where the dynamic face modeling logic operates on every n th frame, and where m and n are integers selected to limit the number of frames on which both the face detection logic and the dynamic face modeling logic operate.

15. The system of claim 10 , wherein the tracking logic further comprises logic to use the set of tracking targets to track the person during their movements within the video feed, the image of the captured member is associated with an appropriate cluster in the dynamic collection without performing the facial recognition logic.

16. A computer implemented method, comprising:
obtaining by a computer, a frame from a video feed;
detecting by the computer, a face within the frame thereby creating a query face;
generating by the computer, a feature vector for the query face;
comparing by the computer, the feature vector for the query face to feature vectors associated with image clusters in a dynamic collection of faces to identify a person associated with the query face wherein the dynamic collection includes a set of tracking targets having members associated with respective image clusters, wherein the image clusters include current features of the person in a series of facial positions and facing directions than a more limited set of images of the person in a fixed image gallery taken at a prior point in time, and wherein images within an image cluster share an attribute for identification with the faces within frames from the video feed using the set of tracking targets;
updating by the computer, an image cluster associated with the person based on the face by adding the query image to

Assignees

• Hewlett Packard Development Co Lp
• Hewlett Packard Development Co Lp

Inventors

• Zhang Tong

Classifications

• G06T7/251

  involving models · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/763

  Non-hierarchical techniques, e.g. based on statistics of modelling distributions · CPC title

• H04N5/77 Primary

  between a recording apparatus and a television camera · CPC title

• G06F18/23211

  with adaptive number of clusters · CPC title