Precise face recognition rapidly pinpoints people of interest in real-time using digital images extracted from video, external image sources and pre-defined watchlists.

The accuracy of face recognition is dependent on many factors including camera location (placement), distance of cameras, resolution (pixels), video quality, lighting, quality of the reference face image, angle of camera, and type of camera and lens.

To achieve the best accuracy for face recognition, cameras should ideally be placed at eye level, there should be good lighting and people should walk and look directly into the camera. However, for most of the scenarios that BriefCam customers encounter, things are a bit more complicated and cameras are not always positioned and configured in an ideal manner for face recognition. For these scenarios we need to figure out how to best balance the variables at hand to achieve the best possible results.

The main market use cases for face recognition are “access control” and “in the wild”.

Access control means that you want to decide whether to let a person enter (access) an area. For this scenario, the person usually identifies themselves and the face recognition is 1:1 (one to one verification), meaning that each face is compared to a reference face in a controlled environment, such as at an airport when passport control compares the passport image in the biometric database with a scan of the person standing in front of them. In this scenario, the accuracy rate is very high since the algorithm is only comparing one face to one identity and since in most access control scenarios, the camera is positioned in an ideal way and other parameters, such as illumination, are controlled and optimal.

“In the wild” uses cases deal with “uncooperative” subjects and is a 1:N (one to many) face recognition that occurs in a non-controlled environment. The 1:N refers to a one-to-many relationship, meaning that each face is compared to many faces in the dataset of images. Here you identify people based on a watchlist. An example of this type of scenario is identifying criminals or thieves in an entrance to an establishment.

The administrator can disable the Face Recognition feature in BriefCam

BriefCam does not store personal information on individuals either by itself or through its users.

BriefCam's software is a GDPR-friendly product, whereby BriefCam can be configured to let approved users view, export and delete data on individuals that is stored in your systems.

For more information, see the BriefCam Data Protection White Paper.

When performing facial recognition based on deep learning and deep neural networks, there is a significance to the makeup of the datasets used to train these neural networks. Specifically, it is important for the datasets to be diverse and varied – in order for the networks to generalize better to different scenarios and provide consistent accuracy on various different subjects.

BriefCam’s face matching engine is trained on datasets that include a variety of ethnicities and representation of both genders.

In addition, the datasets include a variety of scenes from images of constrained face capture – to “in the wild” images that represent surveillance scenarios.

For increased accuracy, it is important to ensure best practices for lighting, video quality, camera position and placement to ensure that the reflection of light against the range of skin tones reduces shadows, reflection and other inaccuracies in face matching capabilities.