Documentation for Axxon Next 4.5.0 - 4.5.10. Documentation for other versions of Axxon Next is available too.

Previous page Face detection tool  Camera requirements for facial detection Next page

The Face detection basic object is triggered every time a face is captured in the frame. Basic object is enough to perform a search for faces in the archive (see Face search).

In addition, the following types of Face Detection tools detection tools based on metadata from the face detection (see General information on metadata) are available in Axxon Next:

  1. Appearance in area – a detection tool is triggered by the appearance of an object and subsequent face capture in FoV.
  2. Loitering in area –  a detection tool triggered by the lengthy presence of an object  and its face capture  in FoV.

  3. Face recognition evasion detection - triggers when anyone in scene wears dark glasses or covering masks, or uses other masking tricks.

The Axxon Next database stores all faces in binary form:

  1. All captured facial images are vectorized* and stored in the t_face_vector table, and their corresponding capture events are stored in the t_json_event table.
  2. Reference images (see Faces) are stored in the t_face_listed table.

* Facial vector is the mathematical representation of a facial image created upon face capture.

Note

These detection tools require Addon Face Recognition Pack to be installed (see Installing DetectorPack add-ons).

Attention!

With an increase in the number of faces in the database, the statistical error increases: the more faces in the database, the more often similar faces will be recognized when searching in the archive. Accordingly, the degree of similarity when comparing the reference face with the captured face will decrease.

This statistical error is relevant if:

  1. The Camera requirements for facial detection are met.
  2. The database contains over a million faces.

An example of the results of calculating the error:

  1. Tevian module, mugshot dataset (good quality photo), 12 million faces in database, and false matching probability is 0.003%. With these initial data, the researchers obtained an identification error of 0.76%.
  2. VisionLabs module, the initial data are the same. The identification error is 0.81%.
  • No labels