Adding Face Recognition to the Nationwide APFIS Biometric System

Papillon Systems is a vertically integrated provider of biometric solutions such as an Automated Fingerprint Identification Systems (AFIS). In twenty years Papillon Systems supplied biometric technologies (fingerprint, palmprint, iris) for identification systems of a national scale to the governments of India, Turkey, Russia, and more than ten other countries.

website: papillonsystems.com

In 2017, the trend of replacing the fingerprint systems with the Multibiometric ones became evident. So, it became necessary to extend the biometric modalities of Papillon's flagship product, AFIS, to face recognition capability. Upgraded AFIS should be capable of supporting two main use cases:

1. Expert tool - non real-time database search of suspects by a face image;
2. Real-time face recognition in video streams from stationary video surveillance cameras and AR glasses.

Due to the significant complexity of the project (AFIS databases may contain 150+ million faces), a face recognition engine with industry leading performance was necessary for this task.

3DiVi algorithm was finally selected due to the combination of good performance, cross platform support, existence of turn-key real-time video processing components, and a flexible licensing option (no-royalties).

To add face recognition to AFIS using 3DiVi products, the following steps were implemented by the Papillon engineers:

• face registration - face capture in accordance with the requirements of ISO/IEC 19794-5 using special photo complex (to enable uniform lighting conditions) or a web camera;
• search for duplicates in the database during the registration;
• search for unknown persons in the database;
• 1:N real-time search mode using workstations;
• 1:N real-time search mode using mobile terminals (Android smartphone, AR glasses);
• ID verification mode (1:1);
• video archive management.

Basic features were implemented by means of low-level libraries provided in 3DiVi Face SDK.

The task relating to real-time video processing was done with the help of the turn-key modules (Face Machine Client frontend module for real-time video processing and Face Machine Server backend module for face images processing and storage). Those modules are supplied as white label for a complete on-prem integration.

The final system is capable of processing the following sources of video: video file and real-time video camera feeds.

The project of adding face recognition to AFIS took one year and was finished in Q2 2018.