Online face clustering

1. A method for image clustering, comprising:

matching a new image to a representative image of a cluster;

adding the new image to the cluster;

setting the new image as a representative of the cluster with a first time limit; and

setting the new image as a representative of the cluster with a second time limit, responsive to a determination that the new image has matched at least one incoming image during the first time limit.

2. The method of claim 1 , further comprising determining a similarity score for an incoming image by a comparison between the incoming image and the new image.

3. The method of claim 2 , further comprising determining that the similarity score exceeds a similarity threshold to determine that the new image matches the incoming image.

4. The method of claim 1 , wherein the second time limit is longer than the first time limit.

5. The method of claim 1 , further comprising tracking an individual through a space, based on a correspondence between images in the cluster and one or more sensors that captured the images in the cluster.

6. The method of claim 1 , wherein the new image and the incoming images are face images.

7. The method of claim 1 , further comprising:

comparing a third image to representative images of a plurality of clusters to determine a respective similarity score for each cluster of the plurality of clusters;

determining that the third image has an above-threshold similarity score for multiple matched clusters;

merging the multiple matched clusters into a new cluster; and

adding the third image to the new cluster.

8. The method of claim 7 , wherein each cluster of the plurality of clusters has a different number of representative images.

9. The method of claim 7 , wherein the determination that the new image has matched at least one incoming image is based on comparison of a similarity score to a first threshold, wherein determining that the third image has an above-threshold similarity score for multiple matched clusters is based on a second threshold, and wherein the second threshold is greater than the first threshold.

10. The method of claim 1 , further determining that a time limit for the new image has expired and dropping the new image as a representative of the cluster responsive to the expiration of the second time limit.

11. The method of claim 1 , further comprising:

determining that a number of representative images from the cluster has exceeded a maximum number; and

dropping an image as a representative image of the cluster that has a lowest remaining time limit.

12. A system for clustering, comprising:

a hardware processor;

a memory, operatively coupled to the hardware processor, that stores a program that, when executed by the processor, is configured to:

compare a new image to representative images of a plurality of clusters, to determine a respective similarity score for each cluster of the plurality of clusters;

determine that the new image has an above-threshold similarity score for multiple matched clusters;

merge the multiple matched clusters into a new cluster; and

add the new image to the new cluster; and

set the new image with a first time limit as the representative image of the new cluster.

13. The system of claim 12 , wherein the new image and the representative images are face images.

14. The system of claim 12 , wherein the program is further configured to set the new image as a representative image of the new cluster.

15. The system of claim 12 , wherein the program is further configured to determine that the new image matches one or more incoming images during the predetermined first time limit.

16. The system of claim 15 , wherein the program is further configured to set the new image with a second time limit as the representative image.

17. The system of claim 16 , wherein the second time limit is longer than the first time limit.

18. The system of claim 12 , further comprising an analysis security console configured to track an individual through a space, based on a correspondence between images in a cluster of the plurality of clusters and one or more sensors that captured the images in the cluster.

19. The system of claim 12 , further comprising an analysis console configured to track an individual's path through an environment that is monitored by multiple video streams, based on a cluster of images associated with the individual, and to perform a security action responsive to the individual's path.

20. A method for image clustering, comprising:

comparing a new image to representative images of a plurality of clusters, to determine a respective similarity score for each cluster of the plurality of clusters;

determining that the new image has an above-threshold similarity score for multiple matched clusters;

merging the multiple matched clusters into a new cluster;

adding the new image to the new cluster; and

setting the new image with a first time limit as the representative image of the new cluster.