Running Face Recognition at the Edge has many benefits. These are:
With Dumb Motion Detection, every motion clip is uploaded to the cloud where the AI is running. This means shorter battery life, costly cloud processing, and high latency. Privacy issues with people’s personally identifiable information stored in the cloud are also a significant risk.
The benefit of running Face Recognition AI on the Edge (instead of the cloud) is that private and sensitive data isn’t sent to the cloud. Cloud processing bills go to $0, and you get faster alerts since it’s all being processed on the camera itself.
Since you no longer need to upload every Dumb motion clip, you can also significantly prolong battery life.
Face recognition at the Edge provides the best data. It allows for high quality images that can be cropped and can get higher resolution faces from part of a scene. This is not possible off-device or in the cloud with transmitted and compressed video.
Face Recognition at the Edge also increases robustness. In an application, identity determined from Face Recognition usually drives a decision or action. Having this technology at the Edge means that the system can continue to work even if the network is down.
The benefits of an efficient architecture design can be seen in terms of better scalability. As a distributed compute model, Edge Face Recognition provides the benefit of not increasing server compute loading with the growth of more devices.
The Edge also serves as an intelligent gatekeeper for reducing video traffic to the server. Additionally, it can redirect server compute resources to more advanced analytics.
The issue with past approaches is that motion detection generates lots of false alarms. We refer to this as Dumb Motion Detection.
Smart Motion Detection works by processing each and every motion clip using an AI model to determine if a person was in that clip. If so, then the clip along with the alert saying “Person Detected” is sent to the user. If not, that clip can be discarded as a false alarm.
Face Recognition is when you search a database of known faces to see if a detected face matches one of those faces. If it matches, that identity and name is assigned to the face. If no match is found, it will be tagged as an UNKNOWN face.
When combined with person detection, Face Recognition allows you to tell familiar faces from strangers’ faces. This is an attractive selling feature compared to standard person detection.
Face recognition allows you to recognize a person’s face so that you can see who’s at your property. Face Recognition can differentiate between family members, frequent visitors, and unknown faces to keep you and your family safe.
See What Matters Using the World’s Fastest Edge Computer Vision AI
Learn more ➔© 2024 Xailient
By downloading or using the Xailient SDK (the “SDK”) or using the Xailient Cloud
Platform, you and, if applicable, the company or entity that you represent (collectively, “you” or “your”) are consenting to
be bound by and are becoming a party to this SDK License Agreement (“Agreement”) with Xailient, Inc.
(“Xailient”). If you are entering into this Agreement on behalf of a company or other entity, you hereby
represent and warrant that you are authorized and lawfully able to bind such company or entity that
you represent to this Agreement. If you do not have such authority or do not agree to all of the terms
of this Agreement, you may not download or use the SDK.
Agreement relates. Any breach of this Agreement by you would cause irreparable injury to Xailient for
which no adequate remedy at law exists, and you agree that equitable remedies, including injunctive relief
and specific performance, are appropriate remedies to redress any such breach or threatened breach, in
addition to all other remedies available. As used herein, “including” means “including without limitation”.
Schedule A: Data Sharing Agreement
Xailient provides a computer vision platform that incorporates machine learning techniques. To take full
advantage of Xailient technology, customers need to train models specific to them. This requires
customers to provide Xailient with their data. Xailient recognizes that data is a critical asset of our
customers, and is being shared with us for the specific purpose of providing the Customer with service.
Xailient uses a combination of Customer Training Data and Xailient Training Data fed into a Training
Process to create Customer Training Output.
Xailient uses the Customer Training Output and other systems to deliver Xailient Products and/or
Services. After training, the use of Xailient Products and/or Services takes place in Runtime. Use by the
Customer may create Customer Runtime Data.
The following are authorizations granted to Xailient and restrictions placed on Xailient:
1. Xailient will use Customer Training Data to train algorithms (Customer Training Outputs) that are
delivered to the Customer or are run by Xailient to deliver services to the Customer.
2. Xailient will not deliver Customer Training Outputs to third parties, nor use Customer Training
Outputs to provide Xailient Products and/or Services to third parties without the express written
consent of the Customer.
3. Xailient will not use Customer Training Data for any purpose other than to deliver products or
services to the Customer, however,
4. Xailient may use meta-data about the Customer Training Data to improve Xailient Products
and/or Services, or to deliver service to the Customer. This meta-data about Customer Training
Data includes but is not limited to:
1. Information about the size and number of images used in the Training Process,
2. Information about the number of classes (object types) and number of instances of
those classes used in the Training Process,
3. Information about the geometry of ground truth instances, such as size, overlap or
intersection with other ground truth instances,
4. Information about the frequency and correlations of ground truth instances,
5. The problem domain being solved by the Customer,
6. The contrast, brightness, average color depth, or other photometric properties,
7. For the avoidance of doubt, the image bitmap (jpeg, png, mpeg video, video frames etc.)
are not meta-data and are only used to deliver products or services to the Customer.
For example, the following would be an allowed use by Xailient:
Xailient uses the image resolutions, proportionate size of ground-truth
bounding boxes, number of classes, and number of images used to train the
Customer Training Output to estimate the number of images needed to train
a model in a different domain. Xailient then provides a third party with a
recommendation of the number of images they should share with Xailient.
5. Customers may opt-in to the collection of run-time images from the customer by Xailient.
Xailient will not use such Customer Runtime Data for any purpose other than to deliver products or
services to the Customer, for example to provide quality control monitoring to the Customer.
6. Xailient does collect meta-data about the Runtime performance of the Xailient Products and/or
Services. This Meta-data about Runtime includes:
a. The number of images or video frames inputted to the Xailient Products and/or Services,
b. The format, encoding, timestamps and series information of the input,
c. The contrast, brightness or average color depth of the input,
d. Statistics about the output of the Xailient Products and/or Services, including number of
frames with and without objects of interest, the frequency of detection, the geometry of
the detection, the size in bytes,
e. The configuration data of the running Xailient Products and/or Services,
f. Information about the Customer’s runtime environment including hardware and
software running in conjunction with Xailient Products and/or Services,
g. Memory, CPU utilization, persistent storage I/O statistics, network connectivity statistics
of the runtime environment,
h. For the avoidance of doubt, the image bitmap (jpeg, png, mpeg video, video frames etc.)
are not meta-data, and are only used to deliver products or services to the Customer (if
collected at all).
7. Xailient may use meta-data about the runtime performance of the Xailient Products and/or
Services to improve Xailient Products and/or Services, or to deliver service to the Customer.
8. Xailient may use meta-data in anonymized, aggregate form for marketing purposes.
For example, the following would be allowed uses by Xailient:
Xailient publically states that on average drone customers need a specified
number of images per class to train a Xailient Detectum™.
Xailient publically states that on average retail analytics customers save a
specified number of bytes per month in wireless data.
The following are restrictions placed on Customers
1. Customer will not install or run Xailient Products or Services without a Xailient License
Agreement.
2. Customer will use Customer Training Output(s) only for the purpose(s) outlined in the Xailient
License Agreement.
3. Customer will not attempt to reverse engineer the Customer Training Output(s) or Xailient
systems or methods.
4. Customer will not attempt to circumvent the Xailient Digital Rights Management or License.
5. Customer will not redistribute the Customer Training Output(s) to any third party except as
outlined in the Xailient License Agreement.
6. Customer will not grant access to Xailient Products or Services, nor provide their Customer
Credentials to any third party.
7. Customer will not use the Xailient technology in any manner that is discriminatory or in violation
of regulation or laws in the jurisdiction in which Customer operates the Xailient SDK or Xailient
Cloud Platform, nor will Customer upload images depicting nudity to the Xailient Cloud Platform.
Schedule B: Platform Entitlements and Usage Limits
Tier: Premium
Maximum Custom Dataset Upload size: 1 GB
Custom Model Training: Selective Attention 5 models per month, per training charge thereafter
Test video upload and inferencing: 20 videos max, 5 minutes maximum duration each
Support: Console support via email for duration of active subscription, SDK support via email for duration
of active subscription, or 1 year for any perpetual licensed SDKs.
SDK: Each SDK license entitles you to process 1 camera. For subscription licensed SDK, the SDK license
may be reallocated to a different camera once per day. For perpetual licensed SDK, the SDK license may
only be reallocated through the reprovisioning process. If a device cannot be connected for
decommissioning, the SDK license cannot be reallocated. A camera is defined as a single CCD sensor, with a
maximum throughput of 30 frames per second.