Introduction

We transform your cameras into intelligent machines capable of identifying any person of interest.

The Face Recognition (FR) feature allows you to identify persons using video streams from your cameras. Recognized faces are compared with the pre-defined database and for every match, a record is added to the archive. This record consists of: time and date of recognition, name of the recognized person, confidence level, camera ID and links to the corresponding video archive files. Using this information, users can quickly review and access specific footage from the end-user portal.

To achieve reliable face recognition, it is required to properly place and configure the camera, enable and adjust appropriate event detection on the camera. Any of the camera modules that generate motion events can be used, and the best option (simple motion detection, line crossing, intrusion, etc.) will depend on the specific scene conditions and camera positioning.

Requirements and recommendations

Step 1: Camera selection

Almost any modern camera can be used for FR, though the best results will be achieved with models that allow you to control image settings such as shutter speed, frame rate, and bit rate manually or semi-automatically (when you can set a range of min and maximum parameters for the camera). A combination of lens and resolution should be such that it ensures that the face will have at least 80 pixels in width at the distance where the person will typically look at the camera. For example, even a 720p camera can be adequate when combined with a narrow lens, while a 4K fisheye camera placed 10 meters away would not be enough.

Step 2: Camera placement considerations

The best camera placement is at an entrance checkpoint or at a doorway where the persons of interest are required to pass in front of the camera. The camera must be installed on the face level or at a small angle slightly higher than a typical person's height in order to ensure the best view of the face. Also, ensure that at least part of the typical path that the person takes when walking near the camera is directed towards the camera; ideally, persons should be looking directly at the camera at least briefly. Positioning attention signs or blinking light bulbs just over the camera can draw attention and ensure that every person looks into the camera.

In the left picture, the camera is at face level, and people coming out of the elevator are looking towards the camera

In the right picture: The camera is positioned too high, the majority of visitors will be passing from side to side, never looking towards the camera

Step 3: Lighting considerations

A person’s face must be adequately illuminated, preventing over or underexposure to the picture. Avoid areas where the person’s face will be illuminated by direct sunlight, or scenes with high contrast, like indoors near big windows, or long corridors with one light source at one end, places with constant changes in illumination due to automatic lights, or upon door opening. Uniform and constant illumination of the faces to be recognized is required for reliable recognition. Avoid bright spotlight sources this will create sharp shadows. Diffused and soft light sources are recommended.
If the camera has to be installed against the light source, a bright object on the background (for example, the sun behind the entrance door), the camera exposure (or brightness) should be set to be above the default (automatic) value, overexposing the background to ensure that the face in the frame becomes lighter and more natural in color.
Also note that some features of the cameras (WDR, Backlight compensation, Highlight compensation, etc.) typically result in blurry frames. Check how these features influence the final image quality, specifically how sharp and contrasting the moving person is. It is better to have under/overexposed areas in the image with sharp face features than even lighting but a blurry face.

Step 4: Camera setup

The camera resolution and the lens must ensure that the person’s face is at least 80 pixels across when passing under the camera, for a more reliable recognition; 150+ pixels is recommended. For low-resolution cameras, this means that the person must pass closer to the camera, or a zoom lens must be used. The training set of images must be at least 150 pixels in the face width without motion blur.
The recommended camera frame rate is between 8 and 15 fps(frames per second).
The camera must have a capability of exposure adjustment to avoid getting blurry images of the passing people. Minimum shutter speed should be limited to 1/60 at most; recommended is 1/100 of a second or faster. The priority between the shutter speed and the aperture must be given to the shutter speed; in some models, this is controlled by the sharpness/smoothness settings, sharpness should be given priority. This means that it is preferable to get a darker image than a blurred one. Depending on the camera manufacturer, this setting can be called “Allow low shutter” or the limitation of the “maximum exposure” set this to 1/60 or less. The cameras without this capability will result in a blurry image that may look better in motion, but the stop-frames will be blurry, decreasing the accuracy of face recognition or even making it impossible to recognize a person. The increased sharpness of the camera leads to significant visual noise, which results in false triggering of the motion detector. To avoid this, provide better illumination at the camera location.

Note: Usage of the H.264 codec with high compression or a large interval between the keyframes may lead to the “tailing effect” of the moving objects in the resulting image. To overcome this, set a higher bitrate in the camera settings and reduce the interval between the keyframes.

Step 5: Prepare all required information to activate Face Recognition

To process your request, we need to know the camera ID and you will need to provide us with the photos of people along with their names or other identifiers that you want us to look after. Photos must comply with the requirements as described in step 4. Ideally, photos should be taken using the same camera under the "normal business" activities, meaning that people should be wearing their normal everyday looks (not a studio photoshoot). Also, to improve accuracy, we will need more than one photo taken from different angles for each person. Make sure that no other faces are present in the photos that you submit, you can either take snapshots when there is no one else present in the frame or simply crop the image and leave just the face and 50-100 pixels around the face, please do not crop too close to the face. For best results person should be looking into the camera on some of the images and being at a slight angle on the other images (left, right, up, and down) making sure that both eyes are visible. Here is an example of a folder with prepared photos, key aspects are:

only one face is in the image
size of the face is over 150 pixels wide (this means that total image width will be 250 pixels or more, just the face should be over 150 pixels in width)
image is clear (no motion blur)
the face is pictured from multiple angles
light conditions are appropriate
each image has a person's name under it.
both eyes should be visible

Step 6. Submit your request

Use the "Submit a ticket" form at the top of this page to create a request to add Face Recognition to your camera. Include Camera ID and images of persons of interest in your request. Make sure your images satisfy all requirements from step 5. You can archive the folder with images to upload all of them at once.

Examples of good and bad camera installations:

For more information on how to optimize image parameters, please visit this article↗️.

How to setup face recognition successfully