Distributed facial recognition, registration, and authentication
Facilitating secure and updated user profile information within a distributed network for facial recognition-based access control systems is disclosed. A user's signature (e.g., facial profile data) is routinely and consistently updated across the access control platform. A remote computing system maintains timely and accurate signature data between user devices and access control devices such that each device benefits from the most updated user information, facial recognition models, and spoof detection models. A user may enroll or disenroll in the access control platform via his or her own mobile device, a cloud-based server, and/or the access control device. The enrollment or disenrollment data may be automatically synchronized and updated throughout the platform. These and other benefits may improve the ease with which users and access controllers (e.g., building owners, landlords, etc.) can securely and automatically enroll/disenroll users, thereby improving the uptake and prevalence of facial recognition-based access control platforms.
1 . A system for facial recognition-based access control, comprising:
one or more on premise access controllers comprising a respective camera;
a cloud access system comprising one or more remote computing systems configured to maintain signature data between the one or more remote
computing systems and the one or more on premise access controllers; and
a machine learning (ML) engine, wherein
the one or more remote computing systems are configured to:
obtain facial images of an individual at different angles,
generate a face signature for the individual using the obtained facial images, and
sync the face signature with the one or more on premise access controllers,
the one or more on premise access controllers are configured to:
capture one or more images of a face of the individual,
authenticate the individual by comparing the one or more captures facial images to the face signature,
automatically permit entry responsive to the authentication of the one or more captured facial images succeeding, and
automatically deny entry responsive to the authentication of the one or more captured facial image failing, and
the ML engine is configured to perform at least one of:
(a) aligning the one or more captured images of the face of the individual, and
(b) cropping the one or more captured images and performing face recognition signature generation, storage, and lookup to compare the individual with a database of face signatures.
2 . The system of claim 1 , wherein the face signature comprises a complete face scan and photo of the individual, a series of photos of the face of the individual taken from different angles, orientations, and lighting, a unique numeric signature generated through machine learning (ML) to correspond to each face photo, a three dimensional (3D) mesh or depth data of the face of the individual, or any combination thereof.
3 . The system of claim 1 , wherein the one or more on premise access controllers comprise one or more machine learning (ML) models, and
the one or more on premise access controllers are configured to use the one or more ML models configured to perform facial recognition, spoof detection, or both, during the authentication of the individual.
4 . The system of claim 1 , wherein the one or more on premise access controllers are further configured to capture three dimensional (3D) depth data in addition to the one or more captured images of the face of the individual.
5 . The system of claim 1 , wherein responsive to an access controller of the one or more on premise access controllers failing to identify the user, the respective on premise access controller is configured to send the captured one or more images of the face of the individual to the one or more remote computing systems, and
the one or more remote computing systems are configured to authenticate the individual by comparing the one or more captures facial images to the face signature and provide results of the authentication to the respective on premise access controller.
6 . The system of claim 1 , wherein responsive to detecting that an on premise access controller of the one or more on premise access controllers is offline while when trying the sync the face signature with the respective on premise access controller, the one or more remote computing systems are configured to generate a timestamp for the face signature, and
upon coming online, the respective on premise access controller is configured to query the one or more remote computing systems for unsynced signatures, and the one or more remote computing systems are configured to compare the timestamp to a time that the respective on premise access controller came online again and provide the face signature to the respective on premise access controller.
7 . The system of claim 1 , wherein upon bringing an on premise access controller of the one or more on premise access controllers online, the respective on premise access controller is configured to obtain copies of face signatures for users that are stored in the cloud access system.
8 . The system of claim 1 , further comprising:
an admin dashboard configured to:
register users, create user permission groups, assign door access to user permission groups, and assign user permission groups to users in the cloud access system, and
send copies of registered users, mappings of user permission groups, mappings of door access to user groups, and mappings of user to permission groups to the one or more on premise access controllers in real time.
9 . The system of claim 1 , wherein the cloud access system further comprises:
a machine learning (ML) engine is configured to train one or more ML models that perform accurate facial recognition of individuals using photos, perform facial recognition based on three dimensional (3D) depth data, perform facial recognition based on infrared images, perform spoof detection by checking whether an image is of a real person or an artificial source, or any combination thereof.
10 . The system of claim 9 , wherein the ML engine is configured to use newly captured facial image data, spoofed image data, or both, to train a new version of a facial recognition model and/or a spoof detection model, and responsive to when an accuracy of the new version of the facial recognition model or the spoof detection model exceeding exceeds an accuracy of a currently deployed facial recognition model or the spoof detection model,
the ML engine is configured to deploy the new version of the facial recognition model or the spoof detection model to the one or more on premise access controllers.
11 . The system of claim 1 , further comprising:
a mobile computing device comprising a camera and an access control application, wherein the access control application is configured to capture the one or more facial images of the individual obtained by the one or more remote computing systems and upload the one or more images to the cloud access system.
12 . The system of claim 11 , wherein the one or more remote computing systems are configured to:
perform an image quality check to ensure that the one or more obtained images meet minimum quality criteria, the minimum quality criteria comprising a size of the face, pitch, yaw, roll, and/or orientation of the face, lighting conditions, image size, image quality in terms of clarity and noise, any combination thereof, in the one or more captured images, and
responsive to the obtained image not meeting the minimum quality criteria, notify the access control application of the mobile computing device that the one or more obtained images do not meet the minimum quality criteria.
13 . An on premise access controller, comprising:
a camera; memory storing computer program instructions; and
at least one processor configured to execute the computer program instructions, wherein the computer program instructions are configured to cause the at least one processor to:
sync face signatures with one or more remote computing systems of an access control system,
capture one or more images and three dimensional (3D) depth data of a face of an individual,
perform at least one of: (a) aligning the one or more captured images of the face of the individual; and (b) cropping the one or more captured images and performing face recognition signature generation, storage, and lookup to compare the individual with a database of face signatures,
authenticate the individual by comparing the one or more captures facial images and the 3D depth data to the synced face signatures from the access control system,
automatically permit entry responsive to the authentication of the one or more captured facial images succeeding, and
automatically deny entry responsive to the authentication of the one or more captured facial image failing, wherein the synced face signatures comprise a complete face scan and photo of the respective individual,
a series of photos of the face of the respective individual taken from different angles, orientations, and lighting, a unique numeric signature generated through machine learning (ML) to correspond to each face photo, a 3D mesh of the face of the respective individual, or any combination thereof.
14 . The on premise access controller of claim 13 , wherein responsive to failing to identify the user, the on premise access controller is configured to:
send the captured one or more images and the 3D depth data of the face of the individual to the one or more remote computing systems; and receive results of authentication by the one or more remote computing systems.
15 . The on premise access controller of claim 13 , wherein after when the on premise access controller has been offline, upon coming online, the on premise access controller is configured to:
query the one or more remote computing systems for unsynced signatures;
received the unsynced face signatures from the one or more remote computing systems; and
store the one or more unsynced face signatures.
16 . A cloud access system, comprising: one or more remote computing systems configured to maintain signature data between the one or more remote computing systems and one or more on premise access controllers; and
a machine learning (ML) engine configured to train one or more ML models that perform accurate facial recognition of individuals using photos, perform facial recognition based on three dimensional (3D) depth data, perform facial recognition based on infrared images, perform spoof detection by checking whether an image is of a real person or an artificial source, or any combination thereof, wherein
the one or more remote computing systems are configured to:
obtain facial images of an individual at different angles, generate a face signature for the individual using the obtained facial images via the ML engine, and
sync the face signature with the one or more on premise access controllers, and the ML engine is configured to perform at least one of:
(a) aligning the one or more captured images of the face of the individual, and
(b) cropping the one or more captured images and performing face recognition signature generation, storage, and lookup to compare the individual with a database of face signatures.
17 . The cloud access system of claim 16 , wherein the ML engine is configured to use newly captured facial image data, spoofed image data, or both, to train a new version of a facial recognition model and/or a spoof detection model, and
responsive to when an accuracy of the new version of the facial recognition model or the spoof detection model exceeding exceeds an accuracy of a currently deployed facial recognition model or the spoof detection model, the ML engine is configured to deploy the new version of the facial recognition model or the spoof detection model to the one or more on premise access controllers.
18 . The cloud access system of claim 16 , responsive to detecting that an on premise access controller of the one or more on premise access controllers is offline while when trying the sync the face signature with the respective on premise access controller, the one or more remote computing systems are configured to: generate a timestamp for the face signature; and
responsive to a query for unsynced signatures from the respective on premise access controller, compare the timestamp to a time that the respective on premise access controller came online again and provide the face signature to the respective on premise access controller.