Face recognition only relevant for the security industry? The possibilities for facial recognition go much further. But software must be able to recognize a face under all circumstances. Recently CEO, Peter Hoekstra, was interviewed in FWD Magazine. Below are the most important outcomes from that interview. The Dutch version of the FWD magazine article can be found here.
Car seat and mirrors that automatically adjust before you click your seat belt? A television that adjusts the channel offer to who is in front of it? It is all possible, and one glimpse of your face is enough, thanks to the software of 20Face.
Faster and cheaper
Facial recognition software offers enormous opportunities in the ‘convenience branch’ and for hospitality, thinks business brain Peter Hoekstra”. We mainly know facial recognition as a security application,” he says. “But there is so much more. Imagine that in a nursing home the elderly are always given the right amount of medication on the basis of facial recognition”. According to Hoekstra, such applications are not used currently due to the hardware and software used. “Infrared scanners or special cameras sometimes have to be purchased and the systems are often too slow. Existing technology compares an input image with images in the database and processes them all. If you want to use facial recognition more broadly, it must be faster and cheaper”.
Lightweight, low threshold
To achieve this, 20Face is fully committed to intelligent software. Purchasing hardware is not necessary. Hoekstra: “Our software can be applied to the cameras that are already hanging in hotels, on the dashcam in the car, the webcam on a laptop and so on. After all, it is not the resolution of the images that form the bottleneck, but the software analysis. “That focus on software is not new, there are now also phones that you unlock by keeping your face in front. But experience shows that this does not work optimally in all circumstances. In bad light conditions the device does not recognize you and systems can, sometimes, be fooled by holding a printed photo in front of the camera. The software of 20Face, however, has enough of a small piece of face. The system also recognizes Hoekstra when it is placed in front of the lens with a wig, glasses and fake moustache or from a strange angle. And while current systems also perform poorly in low light conditions, the 20Face system still works flawlessly under such difficult conditions.
Landmarking and deeplearning
This increased quality is due to algorithms that recognize landmarks, points on the face next to, for example, the nose and the eyes. A few points are sufficient to
register a unique face pattern and the software recognizes a registered face at lightning speed from those same unique relationships between the landmarks. To ensure that the system really recognizes every face that passes, training is needed: deep learning. Hoekstra: “The system learns from ‘labeled data’: with three million photos of eight thousand faces we ‘fed’ the system. On the basis of all these examples, the software learns what a face is and how you recognize different ones
In addition to application in hospitality, this system also offers opportunities for using face recognition much broader and larger-scale for security. Hoekstra: “An alarm on the door will soon no longer be necessary. A camera in the living room ‘knows’ which faces belong there and which do not. Meanwhile 20Face has started a pilot with football club Heracles. The goal of the collaboration: Ticket sales based on facial recognition and access control and personalized service in the VIP area. Hoekstra: “Ultimately, with the system in the masses of people entering the stadium, Heracles can recognize visitors who, for various reasons, require a ‘special’ approach. VIPs or disabled for example. But we are not that far yet. For that, data from thousands of visitors must be processed in a short time. ”
In the near future, 20Face wants to become a platform where people have their own facial features and can choose for themselves which services that are connected to the platform and make them available. Because not technology but privacy issues are perhaps bigger obstacles to the ambitious roll-out. “Your BMW or the medication system in the nursing home will then check with our permission for our system”. “And in terms of intelligence and speed we hope to be even further by that time, so that we can recognize people with less and less information. Ultimately, it should be possible to recognize someone on the basis of a small piece of eye. ”