Clustering-based person re-identification

1. A computer-implemented method, comprising:

obtaining a plurality of captured images that each include an image of at least one person;

for each of the plurality of captured images, calculating relational metrics indicating a relationship between a person in a respective captured image and persons in each of the other plurality of captured images;

based on the relational metrics, clustering the plurality of captured images into one or more image clusters;

assigning each of the one or more image clusters to an identity of an identity database that is stored in memory of one or more computing devices, wherein assigning each of the one or more image clusters to an identity of an identity database that is stored in memory of one or more computing devices comprises simultaneously assigning each of the one or more image clusters to multiple identities based on co-occurrence probabilities; and

storing at least one sample image associated with each of the one or more image clusters in the identity database.

2. The method of claim 1 , wherein assigning each of the one or more image clusters to an identity of an identity database comprises:

determining that at least one of the one or more image clusters represents an image of a person whose image has already been associated with an existing identity that is currently present in the identity database; and

assigning the at least one of the one or more image clusters to the existing identity of the identity database.

3. The method of claim 1 , wherein assigning each of the one or more image clusters to an identity of an identity database comprises:

determining that at least one of the one or more image clusters represents an image of a person whose image has not yet been associated with an existing identity of the entity database; and

creating an additional identity in the identity database; and

assigning the at least one of the one or more image clusters to the additional identity of the identity database.

4. The method of claim 1 , wherein the relational metrics comprise similarity scores.

5. The method of claim 1 , wherein the relational metrics comprise distance scores.

6. The method of claim 1 , wherein assigning each of the one or more image clusters to an identity of an identity database comprises:

generating a plurality of utility scores for each of the one or more clusters, wherein each utility score represents a relationship between the corresponding image cluster and at least one of an existing identity that is currently present in the identity database or an additional identity not yet added to the identity database.

7. The method of claim 6 , further comprising:

generating each of the utility scores based on an average intra-cluster similarity score and an average inter-cluster similarity score that are each determined based on an assignment of the one or more image clusters to each of at least one of the existing identity and the additional identity.

8. The method of claim 1 , further comprising:

automatically pruning the number of images stored in the identity database based on one or more relational metrics computed between stored images.

9. An apparatus comprising:

at least one network interface;

memory; and

one or more processors configured to:

obtain a plurality of captured images that each include an image of at least one person;

for each of the plurality of captured images, calculate relational metrics indicating a relationship between a person in a respective captured image and persons in each of the other plurality of captured images;

based on the relational metrics, cluster the plurality of captured images into one or more image clusters;

assign each of the one or more image clusters to an identity of an identity database, wherein to assign each of the one or more image clusters to an identity of an identity database, the one or more processors are configured to simultaneously assign each of the one or more image clusters to multiple identities based on co-occurrence probabilities; and

store at least one sample image associated with each of the one or more image clusters in the identity database.

10. The apparatus of claim 9 , wherein to assign each of the one or more image clusters to an identity of an identity database, the one or more processors are configured to:

determine that at least one of the one or more image clusters represents an image of a person whose image has already been associated with an existing identity that is currently present in the identity database; and

assign the at least one of the one or more image clusters to the existing identity of the identity database.

11. The apparatus of claim 9 , wherein to assign each of the one or more image clusters to an identity of an identity database, the one or more processors are configured to:

determine that at least one of the one or more image clusters represents an image of a person whose image has not yet been associated with an existing identity of the entity database; and

create an additional identity in the identity database; and

assign the at least one of the one or more image clusters to the additional identity of the identity database.

12. The apparatus of claim 9 , wherein to assign each of the one or more image clusters to an identity of an identity database, the one or more processors are configured to:

generate a plurality of utility scores for each of the one or more clusters, wherein each utility score represents a relationship between the corresponding image cluster and at least one of an existing identity that is currently present in the identity database or an additional identity not yet added to the identity database.

13. The apparatus of claim 12 , wherein the one or more processors are configured to:

generate each of the utility scores based on an average intra-cluster similarity score and an average inter-cluster similarity score that are each determined based on an assignment of the one or more image clusters to each of at least one of the existing identity and the additional identity.

14. The apparatus of claim 9 , wherein the one or more processors are configured to automatically prune the number of images stored in the identity database based on one or more relational metrics computed between stored images.

15. A system comprising one or more non-transitory computer readable storage media encoded with instructions that, when executed by a processor, cause the processor to:

obtain a plurality of captured images that each include an image of at least one person;

for each of the plurality of captured images, calculate relational metrics indicating a relationship between a person in a respective captured image and persons in each of the other plurality of captured images;

based on the relational metrics, cluster the plurality of captured images into one or more image clusters;

assign each of the one or more image clusters to an identity of an identity database that is stored in memory of one or more computing devices, wherein the instructions operable to assign each of the one or more image clusters to an identity of an identity database that is stored in memory of one or more computing devices comprise instructions operable to simultaneously assign each of the one or more image clusters to multiple identities based on co-occurrence probabilities; and

store at least one sample image associated with each of the one or more image clusters in the identity database.

16. The system of claim 15 , wherein the instructions operable to assign each of the one or more image clusters to an identity of an identity database comprise instructions operable to:

determine that at least one of the one or more image clusters represents an image of a person whose image has already been associated with an existing identity that is currently present in the identity database; and

assign the at least one of the one or more image clusters to the existing identity of the identity database.

17. The system of claim 15 , wherein the instructions operable to assign each of the one or more image clusters to an identity of an identity database that is stored in memory of one or more computing devices comprise instructions operable to:

determine that at least one of the one or more image clusters represents an image of a person whose image has not yet been associated with an existing identity of the entity database; and

create an additional identity in the identity database; and

assign the at least one of the one or more image clusters to the additional identity of the identity database.

18. The system of claim 15 , wherein the instructions operable to assign each of the one or more image clusters to an identity of an identity database that is stored in memory of one or more computing devices comprise instructions operable to generate a plurality of utility scores for each of the one or more clusters, wherein each utility score represents a relationship between the corresponding image cluster and at least one of an existing identity that is currently present in the identity database or an additional identity not yet added to the identity database.

19. The system of claim 15 , wherein the instructions are operable to automatically prune the number of images stored in the identity database based on one or more relational metrics computed between stored images.

20. The system of claim 15 , wherein the relation metrics comprise at least one of similarity scores or distance scores.