6. Example: Integrating with Face Recognition

Face Recognition Overview

Face recognition is one of the most requested video analytics Network Optix receives from our channel, as it addresses a myriad of use cases that enhance the situational awareness of Nx Witness operators – everything from enabling keyless access control to identifying block-listed individuals in casinos.

Generally, it is used to solve three basic usage scenarios:

• To alert operators in real-time (e.g. Passlist alerts) of identified people in the captured video
• To search archived video during forensic analysis of an event (e.g. tracking a face across multiple cameras)
• To automate Actions with the Nx Witness Event Rules Engine (e.g. opening a door for an employee, authorizing usage of a specific machine in a production environment, etc.)

Recommended Face Recognition Features

A Faces Database with a Web-Based Management Interface

All Face Recognition solutions require an independent database and management interface to allow operators to enroll and manage known faces.

• The standard way is to implement the database as an independent service with a web interface for management. In this case, the Plugin requires credentials and URL for the service to connect, stored as Plugin settings.

Person Lists

The integrated Face Recognition solution should allow operators to create and manage Lists of Persons – e.g. Block List, Pass List, Blocked List, Employee List, etc.

Automatic Enrollment of New Faces

Automatically assigning an ID to all detected faces allows Operators to search archived video for persons not yet enrolled or on a list.

• Example: Shoplifting Review — if a person is caught shoplifting, an Operator can quickly run a search across the archived video to look for additional past shoplifting events that may not have been initially recognized.

Enrollment of New Faces from Nx Witness Desktop

Adding a new Person to the Face Recognition solution is one of the most common actions an Operator will need to perform. As such, it's important to have this capability deeply integrated into Nx Witness's user experience.

• Use Context Actions — it is highly recommended that developers use Context Actions in the Plugin to request a username and a list name for the identified person. Using Context Actions allows the Operator to click on a Face in the Objects Tab of the Notifications Panel to open a Context Menu to "Enroll New Person" and be prompted with fields (e.g. name, list) to complete enrollment.

Detected Face Object Format

• Type: Person
• Attributes:
• ID (typically automatically assigned ID in a Face Recognition solution database)
• Name (person's name)
• List Name (name of list or lists person belongs to)
• Any other demographics/behavior/mood/clothes/accessories information

Recommended Face Recognition Events

Use Prolonged Events

If possible, all Events generated by the integrated face recognition system should be prolonged – meaning they should contain start and end timestamps.

• For example, once a person is recognized, the Plugin should activate the Event with a start timestamp, and when the person is no longer recognized, the Plugin should deactivate the event with an end timestamp.

Recommended list of events

• Person Detected — this lets the user configure rules to address situations where unknown persons are detected (e.g. unknown persons in sensitive/restricted areas). This is different from when the person is not recognized, as it can be used immediately – e.g. cloud-based face recognition systems may take a while to return a match.
• Person Recognized — this lets the user configure rules for specific individuals (e.g. open the door for employees during working hours). Send a person’s information using a caption and description of the event.
• Person Not Recognized — this lets the user configure rules for unknown people. This event should only be sent when quality faces have been captured and checked (as image quality changes on a frame-by-frame or camera-by-camera basis, it's not a great idea to send this immediately when a person is detected). Wait until a high-quality frame has been checked against the database to avoid superfluous events.
• Person Recognized in List — This allows the Operators to create List-based rules in Nx Witness's Rules Engine. Use “{List}:{Name}” as caption and “Name:{Name},List:{List}” as description.