How Face Recognition Works in a Crowd

By | Published on

Facial Recognition in a Crowd

face recognition systemsThere are different methods used for matching facial biometrics. The simplest method measures various features of a person’s face, such as the distance between the eyes, or the position of the mouth to the nose.  These geometric measurements or vectors are then coded and stored in a database for later comparison. This type of system is usually used in biometric door access control readers

 

The second method is more complex. It uses IP cameras to capture the full facial image and uses as much information as it can. The software then uses various computer algorithms, including machine learning, to build a set of definition data.  This statistical database increases the reliability of the facial recognition system. The more complicated face recognition algorithm is used to identify a person in a crowd

 

Facial recognition used to identify a person’s face in a crowd is different than biometric access control systems. They have different requirements and challenges. 

 

Door control systems capture the face in a controlled environment. There is a relatively small database of faces to compare. The subsequent comparison is also made in a controlled environment. The lighting conditions and face position are the same for both capture and comparison. 

 

Recognizing a person in a crowd is much more complicated. The faces in a crowd can be oriented differently than the initial picture. The illumination can be different, there could be different facial expressions, and shadows that modify the images detected. The face could be partially obscured by other people of the environment they are in.

How 3D Facial Recognition Works

3D Face Recognition in Crowd

3D facial recognition extends the traditional methods of facial recognition to live-stream accurate capture and identification. In this system, the three-dimensional geometry of the human face is used.

 

There are several techniques for capturing the 3D faces. One method uses multiple sensors to create a 3D model of the face. Another method captures the 2D face and then converts it to a 3D image. This transformational process reduces the cost of capture yet maintains reliability. A standard high-resolution IP camera can be used for capture.  The full frame-rate search process has increased levels of difficulty when compared to a controlled environment. The high-performance algorithm used for 3D identification must work even when there is a cluttered background. Subjects are not only in motion, but the direction they are moving is not optimal for face matching purposes. A sophisticated 3D facial recognition system can identify people even when there is motion blurring. It does this by tracking faces through time. The system operates even when there are busy backgrounds, variations in lighting and the possibility of faces being hidden (occluded) by objects or other people.

 

In the real world, there are multiple people present and each one requires simultaneous detection, tracking, and matching. Faces must be recognized even when there is degradation of facial details because of lossy compression that is present in almost all video transmission and storage.

 

The better 3D-facial recognition systems are constantly learning how to perform better by utilizing convolutional neural networks which allow them to train the algorithms to deliver ever-increasing performance.

IP Camera System Integration

Face Recognition Integration with VMS

An IP camera can be used to capture faces in a crowd. The IP camera must be positioned to capture the faces correctly and to provide a minimum of 30 to 35 pixels across the face. To maximize performance the lighting should be well controlled. Cameras that include good wide dynamic range (WDR) are best when there is a large variation of lighting. When it’s dark, the cameras should have enough low-light sensitivity to operate without introducing a lot of noise.  

 

Some video management software (VMS have options for integrated facial recognition. For example,  Ocularis from OnSSI integrates a high-performance 3D facial recognition system. This allows the security person to be notified when a person of interest is detected. It also allows the recorded video to be searched for all the instances when a certain person has been recorded. The biometric integration with Ocularis utilizes the C2P software. This combination allows the user to see the recorded video along with the information about the identified individual.

Summary of Facial Recognition

Facial recognition is a challenging biometric. It is especially difficult when recognizing a face in a crowd. 3D facial recognition systems are one of the most reliable ways to provide facial recognition in a crowd. An IP camera and new software algorithms have become available that reliably recognize faces in a crowd.

 


If you would like help selecting the right facial recognition system for your application, please contact us at 800-431-1658 in the USA, and at 914-944-3425 everywhere else, or use our contact form.