Low Cost 3D Camera Developed for Facial Capture


Biometric systems using 3D face recognition have been under development for many years. Few are in general use because of major problems including the size, cost and capture speed of the camera and the recognition accuracy.

A Canadian company, VisionSphere Technologies has created a new low-cost facial capture system for use in facial recognition systems.

Each part of the system has its own challenges, but the face capture component determines the robustness of the system.

Here are two examples of 3D images taken from the new system – click the image to see a 3D AVI video.

A 3D camera Demo image
3D Demo 1

The Technology

To generate 3D facial images, the VisionSphere camera uses a structured light pattern. This method has been described in literature for over 20 years. The key to VisionSphere’s success is in their patent pending application of the method. This methodology allows them to capture facial images in high dynamic range, reducing ambient illumination issues.

Image acquisition time is a matter of only 1 or 2 seconds, with another 1 to 2 seconds to save the resulting AVI data file.

How VisionSphere captures the face

In order to construct the most reliable and robust face recognition system possible, the simplest and most effective approach to face processing is used.

Initially, it locates the face and the eyes in the image automatically, using proprietary search algorithms.

Next, the software normalizes and crops the image to provide excellent invariance to lighting, head rotation, facial expression and hairstyle.

The software extracts the required information needed for comparing face images, computing a unique “Holistic Feature Code” for each individual. This code provides a strong discriminating capability for comparing faces with very high confidence rates at fast processing speeds.

How Facial Recognition Works

Second 3D camera Demo image
3D Demo 2

Most face recognition and verification systems carry out the same steps, as follows.

  • First, detect the face, locate key features and capture the details.
  • Second, extract facial descriptors from the captured details.
  • Finally, compare the extracted information to a database containing known data.
  • Issues

    The technical problems with face recognition are caused by differences in head rotation, angle, colour, brightness, hairstyle and facial expression. The quality of the camera and face detection algorithms that locate the face and features determine system reliability.

    These are the front-end elements of the system. If they are not accurate and reliable, then the back end system, which is essentially a database lookup, cannot be accurate.

    The VisionSphere system solves many of the issues on the capture side of this extremely interesting but difficult puzzle. It will be interesting to see the next steps.

    Technology Pilot Program

    Blue Bear Networks in Canada recently licensed the VisionSphere technology for use in a pilot program for Canadian Police, using Sun Microsystems hardware.

    Alan Gray is the Publisher and Editor-in-Chief of NewsBlaze Daily News and other online newspapers. He prefers to edit, rather than write, but sometimes an issue rears it’s head and makes him start pounding the keyboard. Alan has a fascination with making video and video editing, so watch out if he points his Canon 7d in your direction.