Traditional two-dimensional facial recognition systems have been in use for more than four decades. Mature 2D recognition systems are available that achieve low error rates in controlled environments, but are quite sensitive to illumination, pose variation, make-up, and facial expressions.
Today, there are more sophisticated 3D images and systems and also systems that convert 2D images into 3D for more accurate comparisons. The advent of 3D face recognition has made this technology accurate, fast, and as simple as fingerprint recognition. The image capture takes place using very high-resolution cameras – usually 16 MP. The cameras also support Wide Dynamic Range (WDR), allowing high-quality imaging even in low-light and extreme-light conditions. Many image-processing functions are executed on the edge device itself (i.e. the camera).
However, facial recognition solutions still require significant processing at the back-end edge data centre. A good rule of thumb is 20 cameras per server. The images from the surveillance cameras feed into the facial recognition application suite on the edge, which then searches them against a reference database. Given that the face-recognition engine runs at the edge, and the process needs to work in real-time, the flow of data from the cameras to the facial recognition application needs to have the lowest latency possible. The fabric can either be wired or wireless, but the facial recognition engine edge data must be located in close proximity to the camera network.
© 2019 FACIOTICS Inc. All rights reserved