Face recognition🔗
Eocortex allows to use several modules that perform face recognition using a database: Face Recognition (Complete), Face Recognition (Light) and Unique Visitor Counting.
However, it is not possible to use Face Recognition (Light) module with other face recognition modules on one camera at the same time.
The modules ensure high recognition accuracy and can be used together with access control systems at the facilities with high security requirements, for example, at banks or restricted access facilities. Another important use of the modules can be automatic identification of the hotel guests, restaurant customers, and the visitors of other similar enterprises.
Comparison of modules🔗
Capabilities |
Versions |
||
|---|---|---|---|
Face Recognition (Light) |
Face Recognition (Complete) |
Unique Visitor Counting |
|
Identification of faces contained in a database |
√ |
√ |
– |
Identification of people wearing sunglasses, headgear, etc. |
– |
√ |
√ |
Relative recognition accuracy |
Medium |
High |
– |
Number of people in database |
Up to 500 |
Unlimited |
– |
Determination of sex and age |
– |
√ |
√ |
Recognition of emotions |
– |
√ |
√ |
Recognition of faces in archive |
– |
√ |
– |
Reports regarding faces |
√ |
√ |
Detection only |
Reports regarding unique visitors |
– |
– |
√ |
Usage of high-performance video card (GPU) |
+ |
√ |
√ |
Usage of several video cards (GPUs) |
– |
√ |
√ |
√ |
Yes |
– |
No |
+ |
Optionally |
Details🔗
The Face Recognition (Complete), Face Recognition (Light) and Unique Visitor Counting modules determine the uniqueness of a face on the basis of the set of features jointly called the “index”. The modules do not use individual points (dots) of faces when determining the index; instead, the image with the size of 100x100 pixels (using the “points” terminology, a face is recognized using approximately 10 thousand points).
The information required for access control based on the recognized face is stored in the event archive of the server to which the camera that recognized the face is linked.
The Face Recognition (Complete) module can identify masked faces with high accuracy; provided that the database contains samples of these persons without a mask. Also, this module can recognize turned faces; despite the fact that only images of faces looking directly into the camera are entered into the database.
The face database may be located on the same server where the recognition is being performed, or, alternatively, on another server of a unified multiserver video surveillance system.
Each entry in the face database contains the following:
One or more images (screenshots and/or photographs) of a person’s face;
Surname, first name, patronymic of a person that are entered manually and are optional;
Additional information (optional text box);
Group affiliation (optional parameter).
The uniqueness of the entries of the database is determined by the face index. Thus, the database can contain several entries with the coincident surname, first name and patronymic (including the empty field).
It is possible to work with the face database from the Eocortex Client application, or using the API.
Using multiple video cards🔗
Face Recognition (Complete) and Unique Visitor Counting modules use all the suitable video cards installed on the server. Every camera with the recognition module enabled is assigned to one of the video cards.
In a video surveillance system that has many cameras, using multiple video cards on the server allows distributing the load on all video cards. In order to distribute the load on the video cards installed on the same server evenly when working with recognition modules, it is recommended to observe the following conditions:
The number of cameras with modules using video cards bound to the server is a multiple of the number of video cards.
The number of cameras with modules using video cards bound to the server is greater than or equal to the number of video cards.
All video cards on the server must be equal or comparable in performance.
Warning
If the video cards have different performance, the load will not be evenly distributed between them.
Warning
The video card specifications must comply with the requirements listed in the module requirements.
Note
When the number of cameras with the recognition modules enabled exceeds the quantity of video cards used, the cameras will be uniformly distributed among the video cards, as applicable, without considering the characteristics of the video streams coming from the cameras and the performance of the video cards (i.e. an equal number of cameras will be assigned to each video card, wherever possible).
Compatibility with other modules (Complete)🔗
Requires Eocortex motion detector |
Neural Networks |
Compatible with modules |
Incompatible with modules |
|
|---|---|---|---|---|
Standard |
Special |
|||
√ |
√ |
√ |
|
|
1 With simultaneous use, a module selection is available for displaying the analysis results in the Eocortex Client application.
√ |
supported and required for the module to work |
+ |
supported and provides additional features of the module |
– |
not supported or not required for the module to work |
⚠ |
not recommended for use with the current module |
Compatibility with other modules (Light)🔗
Requires Eocortex motion detector |
Neural Networks |
Compatible with modules |
Incompatible with modules |
|
|---|---|---|---|---|
Standard |
Special |
|||
√ |
√ |
√ |
|
|
1 With simultaneous use, a module selection is available for displaying the analysis results in the Eocortex Client application.
√ |
supported and required for the module to work |
+ |
supported and provides additional features of the module |
– |
not supported or not required for the module to work |
⚠ |
not recommended for use with the current module |
Usage🔗
Use of the module in the Eocortex Client application.
Related references
/additional-instructions/analytics-module-troubleshooting