Facial recognition systems can enhance security — but many of them can’t adjust for a pair of sunglasses or a surgical mask. Panasonic’s new facial recognition software algorithms, however, can soon account for both in a program that’s up to five times more accurate than earlier programs. The new face recognition server software, WV-ASF950, can still read faces with sunglasses, or faces turned up to 45 degrees from the camera — and before the end of the year, will also be able to recognize faces obscured by a surgical mask.
According to comparison tests by the National Institute of Standards and Technology (NIST), the software improves facial recognition by up to 500 percent. The deep learning software uses a similarity calculation method that allows the program to recognize faces in scenarios most systems struggle with. Along with faces turned from the camera and sunglasses, the program can still recognize faces from the database that have aged from the image in the server.
While designed for security applications, the software also integrates some of the same technology Panasonic uses in their consumer cameras. An A.I. auto-optimizes the camera’s settings in order to achieve a better exposure for higher-quality video. Along with detecting the amount of light in the scene, the system also works to detect how fast subjects are moving to find those optimal settings.
Artificial intelligence also helps the software actually keep costs down, too, Panasonic says. Instead of sending every image to the server, on-camera processing chooses the best shots and sends only those images to the server. By sending images selectively to the server, the system lowers bandwidth requirements, which Panasonic says lowers the cost of operating and building the system. That same concept also allows 20 cameras to run from one server.
The program can recognize up to 10,000 faces in a database, while larger facilities can use an expansion pack to bring that number up to 30,000.
Panasonic worked with the National University of Singapore on the development of the algorithms. The facial recognition system will launch in July (excluding availability in Japan, which will come one month later) and the ability to recognize faces obscured by a surgical mask will be added to the system before the end of the year.
