How can I self-calibrate the camera for optimal face recognition?


Written by NSoft Vision


Guidelines for choosing the ideal camera

  • Recognized mid-to-high-end camera models, with a minimum of 2MP and preferably a Varifocal lens 2.8 to 12 mm, so adjustments can be easily made if necessary.
  • For the initial setup, a PTZ is highly recommended because the pan, tilt, and zoom pan can all be fine-tuned if needed.
  • Ability to combat external light conditions - for example, an entry door camera in direct sunlight or with a lot of concrete reflection glare may benefit from this.
  • Ability to fight dark situations, like as a gate in a low-light environment.
    • Note that object classification does not benefit from high-end camera models as much as facial recognition does. This means that you can use any IP camera to detect objects such as head detection, body detection, car detection, and so on.

Recommended camera settings for optimal Facial Recognition

  1. The resolution should be set to 1920x1080.
  2. Setting video encoding to H264 is highly desirable. 
  3. The frame rate on the camera should not be set to less than 20 fps.
    Ideally, the fps should be set to 25 or 30.

    (Not all cameras support 1080p@30fps, but if your device supports it, it is recommended that you always use a higher frame rate on cameras where you want to use face recognition features)

Recommended camera angles

If the environment allows for shallower camera tilt angles, take advantage of them because they will produce far better results.

  •   Not recommended: Yellow Zone (30⁰ to 50⁰) and Red Zone (50⁰+)
  •   Recommended: Green Zone (0⁰ to 30⁰)
  •   Angle for most reliable identity identification: (0⁰ to 20⁰)



Recommended camera height

  • The tilt angle is usually good at mounting heights of 2.2 meters or lower.
  • Place the camera in such a way that the people in the scene are front-facing rather than looking down.
  • If you cannot manage to find a good spot at mounting heights of 2.2 meters or lower, don't hesitate to reach out to us and we will find the best solution for you.

Built-in visual helpers for camera calibration

If you are an existing Vision Space administrator, we recommend that you use the built-in feature that can help you recalibrate an existing or calibrate a new camera.
To do this, go to Control Paneltune_FILL0_wght300_GRAD0_opsz20.png→ Cameras→ Select Camera.
In the camera details, make sure the Debug Info option is checked on the AI ​​service tab.

Figure 2 – AI Services

An indicator indicating the height of the face in pixels should be visible next to the person’s body.

For accurate face recognition, position the camera in such a way that people passing through the field of interest always maintain a face height of at least 70 pixels.

Figure 3 – Face height vis option