Topics included in this article:
- Camera Positioning and Zooming
- Saving a Face
- Recent Sightings
- Face Details
- Face Alerts
- Fixing an Incorrect Match
- Not a Face Feedback
- Downloading Face Events
- Upload Faces
- Things to keep in mind while uploading faces
Camera Positioning and Zooming
It's good to have the camera mounted as close to face level as possible and no more than 10 feet away from a person's face at the desired point of capture. If the camera is mounted too high on a ceiling or too far away then the feature will not work properly. Furthermore, if you're using this for unique people counting, we generally recommend mounting above an exit door to capture people as they leave a building. Facing inwards often provides the best lighting whereas if you're pointing towards the door their can be bad back-lighting.
To optimize the capture of faces, make sure to adjust the zoom level of the camera. This will help ensure the camera is getting the best views of faces as possible.
This article will give you more details on the max distance your camera can automatically recognize a face.
If you have any question, talk to your Rhombus representative and they will offer a free consultation.
Once facial recognition is enabled, you'll notice a few things. First, you'll see a "Recent Faces" section on the dashboard that is meant to show you recent faces that any camera with facial recognition enabled has seen. If faces are not showing up, ensure the camera positioning is good and that a license is applied.
Saving a Face
Saving a face means you are building a face profile for that person. To save a face, you simply have to name it.
Find the person you want to name, and click on the "Unnamed" link below the face and entering a name.
Once named, the face will appear in the 'Saved Faces' area within the investigations tab.
If you want to un-save a face, simply erase the name and leave it blank.
Now, if you go into Investigations -> Faces, you'll be able to manage the Face Profiles that are saved for your organization.
The recent sightings tab will show you a timeline of thumbnails for all the faces that were recently seen by any camera with a People License enabled. If a face is unidentified, in other words not recognizes by the system, it will be named 'Unnamed'. If you already have started training your model and have a few face profile built up, those faces might named.
By hovering over a face thumbnail you can view details of the captured face (below):
When selected a popup will appear where you can view where the event took place, at what time, and what camera tagged the event. Additionally, you can quickly view the footage by clicking the "View Footage" button:
To learn how to set this up, please refer to this article here.
Fixing an Incorrect Match
If a face is auto-tagged as the wrong person under the Recent Sighting tab, simply hover over the thumbnail and click the "Not a match" button:
Not a Face Feedback
Fortunately or unfortunately AI is not perfect. For events such as these we have exposed a way for you to let us know that the captured face is not actually a face. Do this improves our AI algorithm and thus offers a better performance for you. To do this simply hover over a face that is not a face and select the "Not a face" button:
Downloading Face Events
In the top right of the faces tab you will find a download button. With this button you can download face events to a CSV to then play around with the data at your leisure.
On the Faces tab of the Investigations page, you will see the “Upload Faces” button on the top right of the screen.
To get the best results, please upload 3-10 photos of each person: driver’s license, agency photo, etc. Ensure that images are sufficiently large in terms of resolution. Faces will be recognized as small as 50 x 50 pixels in image resolutions up to 1920 x 1080. Higher-resolution images require a larger minimum face size. Faces larger than the minimum size provide a more accurate set of facial comparison results.
The file name should be in the following format: First-Last_1.jpg; for example: John-Smith_1.jpg
- The dash between first and last name will be read as a space by the console.
- The 1 after the underscore is the identifier – so 1 is the first photo, 2 would be the second, and so on… they will all index as John Smith.
Once John Smith's photos are uploaded, his face will be recognized as John Smith once seen by the camera. You will not have to re-enter his name.
If John Smith has already been seen by the camera and identified as John Smith, and you would also like to upload images, you can use the image uploader to do so. Capitalization must match – uploading images as John-Smith_1.jpg will index into the pre-named face of John Smith already seen by the camera. Uploading images as john-smith_1.jpg will create another face record for john smith.
Things to keep in mind while uploading faces
- Use an image with a face that is within the recommended range of angles. The pitch should be less than 30 degrees face down and less than 45 degrees face up. The yaw should be less than 45 degrees in either direction.
- Use an image of a face with both eyes open and visible.
- Use an image of a face that is not obscured or tightly cropped. The image should contain the full head and shoulders of the person. It should not be cropped to the face bounding box.
- Avoid items that block the face, such as headbands and masks.
- Use an image of a face that occupies a large proportion of the image. Images where the face occupies a larger portion of the image are matched with greater accuracy.
- Ensure that images are sufficiently large in terms of resolution. Rhombus can recognize faces as small as 50 x 50 pixels in image resolutions up to 1920 x 1080. Higher-resolution images require a larger minimum face size. Faces larger than the minimum size provide a more accurate set of facial comparison results.
- Use color images.
- Use images with flat lighting on the face, as opposed to varied lighting such as shadows.