Automated recognition of faces that change over time
Nodes of systems according to the invention perform facial recognition on persons imaged by their respective camera(s) to permit or deny access to resources secured. Recognition is performed using images of faces of known persons stored in the central KP database. Each node maintains a local database with images, vectors or other representations downloaded from the central database for at least some of those facial images. If a node determines that an image acquired by it of a known individual is a candidate for inclusion in the central database based on the quality of the image and of a face depicted in it and on the similarity between that faces and ones in the local database, it can upload it to the server for further consideration. The server can, in turn, determine whether a candidate ought to, in fact, be placed in the central database based on whether that image represents a “life change” in the respective known person, such as aging, weight gain/loss, changes or addition of eyewear, changes in facial hair, and so forth. If so, the central database is updated by way of addition of the candidate image and at least that portion of the database is redistributed to the nodes for use in further recognition activities.
1 . A system for access control comprising
A. one or more nodes, each of which nodes includes a camera and an associated processing device, and each of which nodes performs facial recognition on individuals imaged by its respective camera and, based thereon, generates messaging and/or control signals to permit or deny access to a resource secured by that node,
B. a server that is in communications coupling with the nodes and that includes a central store of facial images of known individuals,
C. at least a given one of said nodes including a respective local store representing at least some of the facial images in the central store and performing said facial recognition using the images so represented in that respective local store,
D. if such facial recognition identifies a match between an individual in an image acquired by the camera of the given node and a facial image of a said known individual represented in the local store of that node, that node determining whether that camera-acquired image is a candidate to update a facial image of that known individual in the central store and, if so, uploading that camera-acquired image to the server, where such determination is based on numerical or other characterizations of a quality of the camera-acquired image and a face depicted therein,
E. the server selecting a candidate image received from the given node for actual inclusion in the central store to update the facial image of the known individual in that image based on a mathematical relation of values of F 1 , a measure of the quality of that candidate image; F 2 , a measure of the degree of similarity of the face in that candidate image and a said facial image of a known individual in the central store; and F 3 , a measure of the degree to which the face in the candidate image reflects a permissible or expected change in that known individual,
wherein the permissible or expected change in the known individual is a change such as aging, weight gain/loss, changes or addition of eyewear, changes in facial ornamentation such as facial hair, tattoos, and jewelry.
2 . The system of claim 1 , wherein
the selection of step (E) is based on the mathematical relation:
I =( k 1 *F 1 )*( k 2 *F 2 )*( k 3 *F 3 )
where,
F 1 is the measure of the quality of the candidate image;
k 1 is a weighting factor for the quality measure;
F 2 is the measure of the degree of similarity of the face in that candidate image and the said facial image of a known individual in the central store;
k 2 is a weighting factor for the similarity measure;
F 3 is the measure of the degree to which the face in the candidate image reflects the permissible or expected change in that known individual;
k 3 is a weighting factor for the permissible or expected change.
3 . The system of claim 1 , wherein, following selection of a candidate image for actual inclusion in the central store, the server downloads to one or more nodes information representing one or more facial images in the central store for the known individual depicted in the candidate image.
4 . The system of claim 1 , wherein the nodes are situated in a vicinity of the respective resource that is secured by that node, and wherein the server is disposed remotely from at least one of the nodes.
5 . The system of claim 1 , wherein the given node determines whether images acquired by its camera of candidates for inclusion in the central store based on scores of any of the uniformity of lighting of the image, contrast of the image, blur of the image, intensity of the image, pose of a face depicted in the image, whether eyes of the individual depicted in the image are open, distance between the eyes of the individual depicted in the image, whether the individual's face is obstructed, apparent distance from individual depicted in the image to a camera that acquired the image, signal to noise ratio in image, and an expected frequency of use of the system for access control and/or a respective node thereof by the individual.
6 . The system of claim 1 , wherein the given node uploads the camera-acquired image determined to be a candidate to update the central store along with any of (i) scores of qualities of that image, (ii) an identifier associated with a matching entry for that image in the local store and/or of the known individual represented thereby, and/or (iii) a measure of similarity between the face in the image and that of the matching entry in the local store.
7 . The system of claim 1 , wherein the central store is updated to include a said candidate image of a known individual that is selected for inclusion therein, and wherein at least a representation of one or more facial images of that known individual is downloaded to one or more of the nodes.
8 . The system of claim 7 , wherein different numbers of representations of facial images are any of maintained in the central store and downloaded to the one or more nodes for different known individuals.
9 . The system of claim 8 , wherein greater numbers of representations of facial images of known individuals who look similar to one another are any of maintained in the central store and downloaded to the one or more nodes than are maintained or downloaded for knowns individuals who do not look similar to one another.
10 . The system of claim 7 , wherein a said representation downloaded to one or more nodes is a said facial image.
11 . The system of claim 10 , wherein a said representation downloaded to one or more nodes is a vector or other representation generated from a said facial image.
12 . The system of claim 11 , wherein the vector or other representation is generated from multiple facial images of a same known individual.
13 . The system of claim 12 , wherein the vector or other representation is generated from a small number of high-quality facial images of a same known individual.
14 . The system of claim 12 , wherein the vector or other representation is generated from a large number of low-quality facial images of a same known individual, where a contribution to the vector or other representation by each of the low-quality images is based on a weighting factor.
15 . The system of claim 7 , wherein the server downloads to the nodes at least a representation of a candidate image that is added to the central store.
16 . A method for access control comprising
A. performing facial recognition with one or more nodes, each of which includes a camera and an associated processing device, and, based on such facial recognition, generating messaging and/or control signals to permit or deny access to a resource secured by that node,
B. maintaining, in a server that is remote from at least one of the nodes, a central store of facial images of known individuals,
C. performing the facial recognition of step (A) on at last a given one of said nodes using a store that is local that node and that represents at least some of the facial images in the central store,
D. if the facial recognition performed in step (C) identifies a match between an individual in an image acquired by the camera of the given node and a facial image of a said known individual represented in the local store of that node, determining whether that camera-acquired image is a candidate to update a facial image of that known individual in the central store and, if so, uploading that camera-acquired image to the server, where such determination is based on numerical or other characterizations of a quality of the camera-acquired image and a face depicted therein,
E. with the server, selecting a candidate image received from the given node for actual inclusion in the central store to update the facial image of the known individual in that image based on a mathematical relation of values of F 1 , a measure of the quality of that candidate image; F 2 , a measure of the degree of similarity of the face in that candidate image and a said facial image of a known individual in the central store; and F 3 , a measure of the degree to which the face in the candidate image reflects a permissible or expected change in that known individual,
wherein the permissible or expected change in the known individual is a change such as aging, weight gain/loss, changes or addition of eyewear, changes in facial ornamentation such as facial hair, tattoos, and jewelry.
17 . The method of claim 16 , wherein
the selection of step (E) is based on the mathematical relation:
I =( k 1 *F 1 )*( k 2 *F 2 )*( k 3 *F 3 )
where,
F 1 is the measure of the quality of the candidate image;
k 1 is a weighting factor for the quality measure;
F 2 is the measure of the degree of similarity of the face in that candidate image and the said facial image of a known individual in the central store;
k 2 is a weighting factor for the similarity measure;
F 3 is the measure of the degree to which the face in the candidate image reflects the permissible or expected change in that known individual;
k 3 is a weighting factor for the permissible or expected change.