EN
RU

VOCORD FaceControl 3D - Next generation of biometric identification system based on 3D.

Recognition in VOCORD FaceControl 3D system is based on synchronous snapshots made by two stereo cameras fr om different foreshortenings, reconstruction of a 3D-facial model of the person based on these snapshots, and matching reconstructed 3D-model against watch list (reference database), containing 3D-model or plane photos.

New technology resolves the main issue of traditional biometric identification system based on 2D-technologies - sensitivity to the pose angle, facial expressions, heavy make-up and uncontrollable lighting conditions.

Key features of VOCORD FaceControl 3D:

  • Detects and captures faces of people in a video stream and reconstructs 3D-facial models
  • Recognizes faces, matches reconstructed 3D-models against watch list (reference database), containing 3D-models or plane photos
  • Alerts the operator about matches
  • Stores in data archive all photos and 3D-models of recognized persons
  • Supports flexible data searching against archive of stored 3D-models
  • Supports viewing of live video from cameras in real time and broadcast video streams over a network

VOCORD FaceControl 3D like VOCORD FaceControl system, is applied to ensuring public security and access control system.

Its difference from VOCORD FaceControl system and traditional facial recognition systems is basic — higher reliability of recognition and efficiency wh ere traditional technologies are inefficient.

How facial recognition system works?

Two couples synchronized stereo cameras of VOCORD NetCam4 deployed in control zone. These video cameras were custom designed for image recognition systems. Whenever a person passes through the check point the system does series of synchronous photos under different foreshortenings.

Check point of VOCORD FaceControl 3D

Based on these photos, a 3D-facial model is reconstructed, reflecting both shape and texture of the original. Then the system defines steady characteristic signs on the face, which are used for matching and recognition procedures. The 3D-model of the person is reconstructed with very high precision (distance between biometric points — to 0,5 mm). The system is able to reconstruct even small textural surfaces like hair, for example.

3D-facial model with shape and texture

The system recognizes faces "on the fly", in real time — the person shouldn't go slowly or stop before cameras. Reconstruction time is about 0,5 to 0,7 per face.

The system matches 3D-models of the persons against the watch list (reference database of 3D-models or plain photos), verifies for coincidence and notifies operator if coincidence is found.
VOCORD FaceControl 3D can match:

  • 3D-models against 3D-models
  • 3D-models against plane images (2D-photos).

Application areas

Security systems:

  • Transport hubs: airports, terminals, metro stations
  • Places of mass stay of people: stadiums, cinemas, business-centers

Border & Access control:

  • Border passport/visa check-points
  • Control posts and security check-points
Up

Robustness to pose angles and artifacts

The main problem of facial recognition in public places, primarily in transport hubs is failure to force people to “cooperate” with the system. People in a facial recognition zone behave naturally: they laugh, talk, eat, and not always look frontally in relation to the camera. These factors decrease recognition accuracy of traditional facial recognition systems significantly.

Due to innovative 3D-technology, VOCORD FaceControl 3D system is robust to:

  • Pose angles
  • Heavy make-up and “war paint”
  • Artifacts – eyeglasses, scarf, moustache, beard, cigarette, etc.
  • Facial expressions (smiling, frowning, yawning, chewing)

fc3d03.jpg

Robustness to variations of lighting conditions

In contrast to traditional facial recognition systems, VOCORD FaceControl 3D is insensitive to swings in light-and-shade as it analyzes a 3D- model rather than a plane snapshot. Moreover, it can work in daylight (solar) illumination conditions and requires no special light sources like structured illumination.

2D + 3D = 2.5D

VOCORD FaceControl 3D system uses not only 3D but also 2D recognition algorithms. If the database of 2D/3D images contains only standard (plane) photos, VOCORD FaceControl 3D system uses the special “2.5D-facial recognition” algorithm.

The system creates a standard 3D-facial model, then makes corrective turn of the 3D-model to the frontal pose and synthesizes frontal 2D-image. Then the system matches the synthesized 2D-image against real 2D-portraits in the database. The algorithm allows of using 3D-modeling benefits – robustness to angle pose, normal illumination, even though the database of 2D/3D reference database contains only plane photos.

fc3d02.jpg

System can distinguish between an human being and facial photo

Traditional facial recognition systems could not distinguish between person and his/her photo. VOCORD FaceControl 3D system analyzes a 3D-model of the object and differentiate the real person and the photo. Moreover, the system is capable to analyze natural facial micro motions, so it cannot be “deceived” even in the case of the sculpture of the head.

VOCORD FaceControl 3D has a geographically distributed architecture. In case of a geographically distributed system, each software module can be installed on a separate server, and there may be an unlimited number of servers in the installation network.

fc3d01.jpg

1. VOCORD NetCam4
Two pairs of high resolution VOCORD NetCam4 cameras. They make 4 synchronous photos of a person to reconstruct facial 3D-model.

2. FC Catcher

Software module capturing facial images from a video stream. Up to 4 stereo cameras can be connected to each server equipped with FC Catcher.

3. FC Grinder

Software module defining biometric parameters of captured face images and calculating the biometric templates.

4. FC Matcher

Software module which matches biometric templates with images in the watch list and decides whether two images coincided and what is the recognition accuracy.

5. FC Archive

Long-term archiving software module. It stores all images, 3D models, biometric templates, system logs, etc.

6. FC Client

Software module for an operator’s workstation. It allows the operator to view live video images from different check points, receiving successful face recognition alerts as well as performing archive retrieval of video data.

Parameter

Value

Architecture

Geographically distributed

Communication interface

TCP/IP

Video cameras

Paired stereo cameras VOCORD NetCam4 (1.4 MP, 20 fps)

Illumination requirements

At least 300 lx

Server requirements

nVidia Cuda video cards is required

3D-model reconstruction speed

From 0.5 to 0.7 s

Biometric template calculation rate

20 ms (CUDA); 200 ms (CPU).

Reference database size

Up to 10 000 faces

Algorithms

3 types of algorithms: 3D (shape matching), 2D (texture matching), 2.5D (matching of 3D model against standard (plane) photo)


3D-face modeling from snapshots
Loading Player
VOCORD FaceControl 3D. Product video
Loading Player
3D facial recognition system VOCORD FaceControl 3D
Loading Player