Face Recognition

If you have installed the video analytics module on the streamer, then face detection and recognition will be available in Flussonic Watcher for the cameras on that streamer.

Face detection is finding a face on a video frame. Face recognition is matching detected faces with the person (face sample) database to answer the question "Who is it?".

When you enable the face recognition function on the camera, Flussonic Watcher starts detecting faces; recognition of the detected faces starts as soon as you populate the person database.

Detection and recognition serve different purposes.

The face detection mode is useful if you need to:

• Eliminate false triggering of the motion detector on foliage, animals, or other moving objects.
• Accumulate a person database, which can later be used for face recognition.
• Get statistical information about the passages of unique faces under the camera.

Face recognition is used as part of integration solutions to solve various problems including:

• Arranging access without card swiping in an access control system
• Employee time&attendance tracking
• Accounting for incoming/outgoing traffic of people

• Automatic identification or verification of persons when performing various actions

  Note

  Flussonic Watcher does not support the above face recognition solutions by itself. You will have to use the API provided to integrate with a third party system(s) of your choice in order to use them.

Content:

• Recommended video image parameters for face recognition
• Enabling face recognition
• Face detection
• Face recognition

Recommended video image parameters for face recognition

A general requirement for a video image for face recognition is the ability to recognize the face with human eyes. In other words, if you do not perceive the face in the video, then Watcher will not be able to recognize it either.

Stable face recognition is guaranteed with the following video image characteristics:

1. No more than +/- 20° vertical and horizontal deviation from a right-angle (90°) view into the camera, i.e. camera should be installed at a height of no more than 2 m.
2. The shutter speed is not more than 1/100 (for example, the face is blurred when moving at a shutter speed of 1/25).
3. Frame resolution at least 1280x720 (720p). If the frame has larger resolution, it will be reduced to 720p before feeding to analytics module. Also, if camera has several streams, the closest to 720p will be used for analytics purposes.
4. The face height should be at least 1/6 of the frame height at the resolution of 1280x720. For higher resolutions, the face size should be larger in proportion to the increase in frame size. In other words, resolution should be 500 pixels per 1 m, i.e. the distance between the pupils must be at least 50-60 pixels at 720p.
5. Lighting not less than 150 lux with face evenly illuminated.

Example of proper camera position

Example of bad camera position

However, if your video image does not meet these guidelines, you can check to see if recognition will work. If the face recognition quality turns out to be unsatisfactory, we may consider the module modification to suit your conditions. Please contact our technical support team by following these instructions. Technical support team member will request the necessary information (for example, access to the stream you use for recognition) and will let you know if the improvements are possible and when we will implement them.

Turning on face detection

Enabling video analytics on the server

By default, the analytics functions are disabled on all servers in your system. To enable analytics:

1. Go to Health and click the streamer name.

2. In the Vision list, select which analytics module you plan to use: for CUDA GPU or for CPU.

   

3. Click Save.

Setting up face detection on the camera

To start faces detection and recognition:

1. In the Watcher UI, go to Cameras. Find the camera in the list and open its settings by clicking the icon in the upper right corner of the player.

2. Check the Enable recognition box.

   

3. Select Recognize faces in the drop-down list that appears when you check the box.

4. Select the period to store precise thumbnails for, if necessary. Please refer to Camera settings for details on precise thumbnails.

   Note

   The recognition settings may not be available if a non-adjustable preset is selected. In this case, select or create the preset with the settings you need.

5. By default, the recognition system searches for faces over all the camera field of view. You can select specific polygonal area(s) to detect faces in by clicking Set up the detection zone. This settings may help you to reduce false detections. The areas are set in a dialog:

   

   There are controls in the dialog header: buttons for selecting already created areas, as well as buttons for creating and deleting areas.

   You can drag-and-drop the created area to the desired position on the frame and move the vertexes. If you want to delete an area, select it and click the trash can icon. Click OK when all areas are set.

   Note

   Make sure that your areas are convex and do not intersect each other. Any area should cover at least quarter of the frame. You will see a warning if any of the requirements is not fulfilled while the OK button will be disabled until you make fixes according to the remarks.

   Note

   Areas are set individually for each camera, even if you select a non-configurable preset.

Face detection

Face detection starts when you turn on face recognition on the camera. All detected faces will fall into the Face detector tab in the Events section. The following information will be displayed: a photo of the face at the time of detection, the date and time of detection, the name of the person (if matched a sample, i.e. recognized). You can also view or upload a video with the process of the person passing before the camera.

Click the event to view its recording. The player will appear on the right, playback position will be set to the recording (maybe a few seconds before the detection itself). You can control playback just like in the usual player.

See also Events.

Face recognition

For face recognition to work, you should populate the person database in Watcher by creating persons and lists of persons. A person is a sample face saved in the Watcher database with accompanying attributes, such as ID, name, etc. A list of persons is a set of persons united on some logical basis, for example, people from the same department or having the same access level to the premises. Each camera can recognize persons from one person list only; you should take it into account when creating lists of persons.

Flussonic Watcher supports face recognition in 1:N identification mode, that is, it allows you to find out which person corresponds to the face detected in the frame.

The following user permissions are required to manage persons and person lists:

• Watcher Administrator can view, edit, or delete all persons and person lists in the system; OR
• Organization Administrator (owner) can view, create, edit, or delete only the persons and person lists within the owned Organization (first of them if many).

You can create one or several person list(s) and add several persons to them. As the persons pass under the camera, events about their passages will appear in the system on the Events -> Face detector tab. If a face is recognized, then the person’s name will be indicated in the event, and if the face is detected but not recognized, then you will see an identifier of a newly created unknown person. You can edit such persons later as described above.

With API, you can use the face recognition events in other systems (for example, when integrating with access control systems, when you want to allow only employees of a particular office to enter the door).

View, edit or create persons

To view the existing persons, go to the Events -> Face detector section and click the Person lists button.

The lists of all persons that are created in the system and permitted to the current user will be opened, as well as a list of persons found on the video that do not belong to any list.

Use the filters on the top of the list to search for person by name, or filter by Organization or person list.

Select Edit in the menu on the right to change the person's data. Click Add person in the top of the page to create a new person. The forms to create or edit persons are the same:

Fill in the form:

1. Type the person's Name.
2. Select the Organization to which the person shall belong.
3. Select the Person list to which you want to add the person.
4. Upload the person's reference Photo, which will be used to compare it with all people passing by, then specify which list of people the person will belong to, and also add an arbitrary note.
5. Optional: Enter the External ID which is used for integration with Access Control System or other third party system. At such integrations, Watcher usually have to transmit message about the person it has recognized. However, person databases in Watcher and external system are not related whatsoever. This is why you should enter the external ID of each person i.e. the ID person has in the external database. This will help the external system to understand which person the message from Watcher is related to and proceed with granting/denying access according to its internal logic or performing other actions.
6. Optional: Type the text Note about the person. The note may include any useful info about the person.

View, edit or create person lists

Click Edit next to the Person list field on the Events -> Face detector -> Person lists page.

You will see the person lists already created in the system.

You can delete the list using the menu. In this case, the persons from the list will be included in the Persons not on any list list.

The menu also has an option to view and download a report on visitors.

To create a list, click Create at the top of the page. To edit the list, select Edit from the menu. Forms for creating and editing lists are the same:

Fill in the form:

1. Type the person list Name.
2. Select the Organization to which the person list shall belong.

3. Select the Camera on the video from which you want to recognize persons from the list. One camera corresponds to one face list.

   Both recognized and unrecognized persons are included in the lists linked to the corresponding cameras. If a face is detected or recognized on a camera that is not linked to any list, then the person is included in the Persons not on any list list.