Managing Facial Recognition

Topics included in this article:

• Overview
• Camera Positioning and Zooming
  • Best Practices
• Search Faces
  • Navigate to Search Faces
  • Search and Filter
  • Matched Face Event
  • Unmatched Face Event
• Manage Faces
  • Navigate to Manage Faces
  • Photo Selection Best Practices
  • Add a Person
  • Bulk Add People
  • Edit a Person
    • Edit a Name
    • Change Matchmaker Photos
  • Delete a Person
  • Add Labels
  • Matchmaker Images
  • AI Matching Threshold
• Face Alert Policy using Labels
• Helpful Links
• Contact Support or Sales

Overview

Rhombus Facial Recognition helps organizations identify and search for individuals captured by supported cameras using AI-powered face matching. This article covers the recommended camera positioning and environmental setup needed for reliable facial recognition performance, along with how to search face events, manage saved people and matchmaker images, configure AI matching thresholds, and create face alert policies using labels.

Camera Positioning and Zooming

Lighting, camera angles, and camera positioning all play critical roles in establishing an optimal facial recognition environment. An inadequate setup in any one of these areas may result in missed or inaccurate facial recognition.

Best Practices

• Mounting Position: Place the camera as close to face level as possible with an angle under 30° from all directions (up, down, left, right).
• Camera Distance: Mount at a distance that provides approximately 100 PPF (pixels per foot) on the face at the desired capture point. For maximum distance guidelines, refer to our article on Optics and Object Distances for Analytics. Using shorter distances than the maximum listed can help reduce bandwidth consumption.
• Lighting: Ensure the camera faces inward to avoid back-lighting and is placed in a well-lit area. Minimize shadows and glare on the face.
• Focus: Adjust the camera’s zoom to focus on the intended facial capture area.
• Encourage Attention: Position the camera where people will naturally look toward it, such as near a sign or graphic to draw attention.
• Uncooperative Scenarios: For cases where individuals don’t actively engage with the camera, try to optimize face capture angles for better results (i.e., avoid side or downward views). Anything that can be done to get a straight shot of the face will enhance the feature's results.
• Unique People Counting Feature: For unique people counting, Rhombus recommends mounting the camera above an exit door to capture faces as people leave.

Search Faces

Navigate to Search Faces

1. Navigate to the "Investigations" tab, and click "Faces."

2. Click "Search" in the upper left corner.

Search and Filter

Search for a matched face by name and filter results by time frame, location(s), camera(s), or label(s).

Enable the "Show only matched faces" toggle to filter out unmatched faces.

Matched Face Event

Matched face events will include the model’s confidence percentage for the face match, along with the associated name.

When a face is captured by a camera, clicking the ellipsis button provides several options:

• View similar sightings - View other sightings for faces matched to this name profile.
• Add as matchmaker - Add this face capture as a matchmaker for this name profile.
• Not a face - Mark the image captured as not a face.
• Remove name - Remove the face as a match to the current name profile. 
• Click the label icon to add or create a new label to associate with the name profile.
• The name of the match can be edited by clicking the pencil icon beside the name. 

Unmatched Face Event

An unmatched face event is an instance where the confidence threshold for matching the face did not meet the minimum value set in the “AI Matching Threshold” section and, therefore, could not be confidently matched to a person saved in the system.

To edit the event, click the ellipsis button beside "Unmatched." Here, you can either mark the event as "Not a face" or associate it with a person by adding their name. 

Note: Adding a name to an unmatched face event will not affect future detection accuracy for that person. All face matching is determined only by the matchmakers loaded and selected in the "Manage Faces" section.

If the name does not already exist as a saved person, you can create a new profile by typing in their name and clicking "Add Person." This will take you to the "Add a Person" flow. 

Manage Faces

Navigate to Manage Faces

1. Navigate to the "Investigations" tab, and select "Faces."

2. Click "Manage."

Photo Selection Best Practices

For optimal face matching, we recommend following a few best practices when selecting images to use as matchmakers.

• Start with "Upload": We recommend starting with high-quality front views for the uploaded photo and seeing how the system performs before adding in additional matchmakers from captured footage. Only one uploaded photo per person is required. 
• Image Clarity: Use clear images or captures with minimal to no blurring.
• Face Angle: For the "Upload" photo, we recommend using a high-quality front-facing image of the person. For additional "Captures from Footage" images, selecting a variety of small angles will assist the model. It is not recommended to use full side views (90 degrees) or multiples of the same angle.
• Maintain Updates: If a person has significant changes, such as growing a beard or getting a dramatic haircut, we recommend re-uploading an image of that person and updating any additional matchmakers.

Add a Person

1. Click "Add Person" in the upper left corner.

2. Name the person you are adding. Optional: Select images from a camera(s) to add here as matchmakers. If you have not yet uploaded an image, we recommend waiting to select matchmakers until after doing so. You will be able to upload an image once the person's profile has been created.  Click "Save" when finished.

3. Once the person has been created, you can click the "Upload" icon next to their name.

4. Select the image you wish to upload from your computer, and it will populate in the "Uploaded Face" column.  Note: This system can only recognize human faces and is run through a preliminary detection to ensure quality and compatibility with the model.

Bulk Add People

1. Click "Bulk Upload."

2. Image naming is important. If you name an image Jane Smith.jpg, the created person will also include the space in the name Jane Smith. If dashes are used, these will be included in the person's name in the console if creating a new person.  If you want to "Create Person" with the uploaded photos, ensure this toggle is switched on. If the name of the image already matches the name of an existing person, this image will become associated with that profile.  Click "Select Image(s) For Upload" to select all images you wish to upload.  Click "Upload" when complete.

Edit a Person

Edit a Name

1. Click on the pencil icon next to the person's name.

2. Write the new name, and click "Save."

Change Matchmaker Photos

1. Hover over an image and click the "X" to delete it. Confirm the deletion on the next window by clicking "Delete."

2. Click the "+" symbol to add an image to an empty slot.

3. Select an image and click "Save."

Delete a Person

1. Click the trashcan icon next to a person's name.

2. Click "Delete" to confirm the deletion of the person's profile.

Note: Deleting a person's profile will also remove all their matchmaker photos. These photos will no longer be available for viewing when looking in the "Search" section at past event matches.

Add Labels

A label is a logical tag to an object that allows users to group faces and apply an alerting policy to them.

1. Click the label icon under a person's name.

2. Type in the word you wish to use for a label. If that label doesn't already exist, click "Create label [name]." If it exists and is actively being used with another person, you can select it from the dropdown list.  Removing a label from all people it is associated with will permanently remove that label from the dropdown list.

Matchmaker Images

These are the faces that will be used for all facial matching for each person. A person can be associated with a maximum of up to five matchmaker photos at a time, with one being an uploaded image and the others being images captured by a Rhombus camera.

Images that can be selected as matchmakers have already gone through a preliminary check. This check looks for a minimum upload quality and meets a specified confidence level. A low confidence level image will not be available for selection, even if it is a captured image from an event on the search page. 

Matchmakers can be added when creating a new person, or by clicking the "+" symbol in the "Captures From Footage" section after a profile has been made.

AI Matching Threshold

The confidence threshold used by the AI Facial Recognition model can be manually set. The default confidence threshold is 70%, which represents the minimum confidence level required to automatically match a captured face with a named profile. 

Increasing confidence improves AI Matching accuracy but may filter out detections. This may increase the number of unmatched events when searching.

Decreasing confidence reduces AI Matching accuracy, which may result in false positives, or names being associated with the wrong image of a person.

Adjusting confidence doesn't account for environmental factors that impact AI performance. Learn more in the AI Confidence Settings article. 

1. Click the dropdown menu in the "AI Matching Thresholds" section.

2. Select the value you want to set as the new threshold. Note: The lowest available confidence value is 50%.

Face Alert Policy using Labels

Once labels have been created and added to user profiles, they can be used to configure a Face Alert event in a policy. 

1. Follow the steps in Creating and Assigning Alert Policies to begin the policy process and ensure the policy is assigned to the camera(s) using facial recognition.

2. Add "Face Alerts" and select a schedule to use.

3. Scroll to the "Face Alerts" section, select any labels you wish to be alerted for, labels you don't want to be alerted for, and toggle on alerts for unmatched faces if needed.

4. To remove, click the "x."

5. To ensure Users are configured to receive the notifications, reference the User-Based Alert Notifications article.

Helpful Links

• Console Features & Licensing
• Digital Zoom and Pan
• Optics and Object Distances for Analytics
• Using the Investigations Tab
• Face Alerts with Labels
• Creating and Assigning Alert Policies
• AI Confidence Settings
• User-Based Alert Notifications

Contact Support or Sales

Have more questions? Contact Rhombus Support at +1 (877) 746-6797 option 2 or support@rhombus.com.

Interested in learning more? Contact Rhombus Sales at +1 (877) 746-6797 option 1 or sales@rhombus.com.