Facial recognition has really come into its own over the past two years, no doubt helped by the Covid-inspired panic about touching stuff. There were always people who got cold chills when asked to touch a fingerprint reader that others had touched, but Covid made touchless the norm and allowed facial recognition solutions to experience significant growth – both in terms of sales and technology advances.

In the residential estate market, access control and visitor management were never seen as a good market for facial recognition as access is mostly outside and light and weather conditions can impact the image of a person throughout the day and night. And then there are those who refuse to turn their heads and actually look at the camera for some reason, making facial biometrics even more complex.

While this writer would have been one of the naysayers when it comes to facial recognition at estate entrances, I was surprised to hear it already works on many estates. This led Hi-Tech Security Solutions to ask three of the primary facial biometric players in the local market for their input on the viability of facial biometrics ‘at the gate’, including the claims of some that their system can recognise faces through closed car windows.

Nicolas Garcia, regional director of sales for IDEMIA, says the trend today is towards contactless as it provides the ideal combination of hygiene, performance and ease-of-use. Facial recognition usually is the first option that comes to mind, but contactless fingerprint scanning is also a very valid alternative that also ticks all those boxes. IDEMIA has seen successful implementations of both technologies.

“At the end of the day, it all comes down to specific sites’ and users’ preferences: some will prefer to use their hand, while others prefer to present their face. Each biometric has its pros and cons.”

Hendrik Combrinck, MD of ZKTeco South Africa, agrees that it is a great option as the technology has proven itself in the residential market as the ‘perfect touchless option’, as the user doesn’t even have to reach out with his hand to do the authentication. This feature puts the user-friendliness of the technology at a much higher level than any other biometric option in the market now.

Walter Rautenbach, MD of neaMetrics and Suprema SA, is also in agreement. He says that, if it works, it is more convenient as you don’t have to put down your groceries and you don’t have to touch anything, which creates more peace of mind independent of whether spreading the pandemic through touching is really that prominent. Furthermore, it also allows for incorporating temperature reading and mask detection, should you insist on the same for your residential estate.

Throwing light on the biometric

As noted above, the big question when looking at facial biometrics at the gate is light. Whether it’s the changing angle of sunlight throughout the day, the lack of it on a cloudy day and at night, how will the camera on the reader deal with the lighting conditions to deliver a reliable result. In addition, what happens when someone is wearing a mask, or they shave their beard off? And what about the claim that recognition can be done through a closed car window.

“Most facial recognition solutions use traditional 2D cameras, which work like an eye,” explains Garcia. “They become blind if they receive direct sun on their lens, much like when taking a photograph with a phone while the sun is in front of you. The picture is dark and you cannot see the person you photographed. Readers using only 2D cameras cannot circumvent this reality.”

To alleviate that problem, IDEMIA’s VisionPass embeds three cameras (2D, 3D and infrared) to maximise the chance to identify a person. Implementation is key to success. He adds that wearing masks, glasses, headsets or other accessories, should usually not have an impact on the quality of reading and one should not have to lower the mask to be recognised.

With car windows, the situation is more difficult to evaluate as several factors can impact the reading: How dark is the window? Does it have anti-glare coating? Is it clean? “If it might work from time to time, it is probably not reliable enough to ensure the right level of security versus convenience,” says Garcia.

Apart from lighting and masks, he notes that another key selection criterion for facial biometrics is resistance to spoofing attempts, because who would want to invest in terminals that can be easily fooled by simply presenting a picture on paper or on a smartphone screen? The best terminals successfully pass evaluations like the iBeta Presentation Attack Detection (PAD) – including VisionPass.

Combrinck agrees, stating that the environment does play a role in the effectiveness of facial recognition technology. “It all comes down to if the installation of the device was done with the correct specifications with a sun shield (if needed) and was the user enrolled correctly from the start. “Because all facial recognition is based on camera technology it will still be influenced by total direct sunlight, but the advances with true WDR technology in cameras is making it less likely that lighting will play a role.”

The issue of the user being enrolled correctly plays a much bigger role, he adds, stressing that a person can be correctly authenticated with dark glasses and a mask as most facial algorithms are based on the whole face (at the enrolment stage). “Therefore, it is of utmost importance that a user doesn’t wear a mask or glasses when being enrolled so the algorithm can build a full template of the face to then cater for other instances when the face is partially covered.

“We have also seen that facial recognition is possible through a car window depending on the lighting on the face in the car and the angle of the reader to the car window, Combrinck adds. “Most of the successful cases of facial recognition through the car window are when the reader is parallel with the car window and is directly in line with the user’s face.”

Compare apples with apples

When it comes to facial recognition readers it is important to compare apples with apples, says Rautenbach. These readers make use of different types of technology, such as visual face recognition, infrared, 3D modelling, etc., which greatly impacts on their performance in variable light conditions.

With pure visual face recognition, we see camera technology relying on what you can see with the eye, which has a huge impact on where it is implemented. If the sun is too direct or if it is too dark, there will be problems. A further downfall of pure visual face recognition is the susceptibility to faking a face by holding up a photo on a piece of paper or on a mobile phone, although most reliable visual face recognition providers are now using techniques to detect this, which have become very accurate.

With non-visual technologies, where external sensing such as IR is used to get both a picture and build a model of the face, the performance is much better. How much it overcomes bright and dark conditions still depends on the strength of the external sensing technology. Faking this technology is also extremely difficult, he adds.

“With this said, no matter what technology is used we must consider the practicality,” notes Rautenbach. “Clients that want to put a face reader under a downpipe just because it is IP65, or do not consider that certain portions of the technology, like selfie enrolment, will be susceptible to light, or who are not considering that some prefer ultimate privacy through dark tinted windows, is really looking to promote challenges and good recommendations should rather be sought from professional installers.

“Can facial recognition work through car windows or with dark glasses? Yes it can, but if it will always work with effectiveness and without compromising security is the real question to be asked. On additional functions such as mask detection and temperature reading, this should be something which your provider should have an option for.”

The curse or blessing of selfies

When talking about facial recognition as an identity authentication mechanism at estate entrances, looking at the current trend of using selfies for identification – signing up for bank accounts etc. – should they also be considered for visitor management. Self-registration via selfie would make the processing of visitors simpler and hopefully more accurate using facial recognition.

Once again, our interviewees are comfortable with the idea. Not only with the idea but it is already happening. Combrinck says this type of visitor management is used daily by many of ZKTeco’s integration partners where a visitor is onboarded via a mobile app or web app prior to arriving at the site. “Our algorithm was designed with these applications in mind and this, again, is what makes facial recognition such a flexible and user-friendly technology because the onboarding can be done on almost any type of camera including cellphone, tablet, webcam or at the device itself if the resolution standards are adhered to for a good quality picture.”

“Suprema’s latest generation of facial readers support both visual and IR recognition,” explains Rautenbach. “The decision in offering both creates more opportunities, because visual recognition allows application flexibility of remotely enrolling visitors or tenants while securing the recognition with IR during first contact, through a process called Fusion matching, offers the best of both worlds.

“Most new smart devices take sufficient quality images for good results and guidelines are provided for selfie enrolment and if these are adhered to you shouldn’t have any problems. It comes down to enrolment quality. What you put in is what you get out.”

This is already a reality supported by IDEMIA’s technology, although it requires a bit of integration. “We worked on a similar case in the UAE with a partner of ours where the end-user wanted to use VisionPass to manage both their employees and their visitors,” Garcia explains. “In this case, visitors use an app developed by our partner and enrol themselves through a selfie on their phone. The template is then automatically loaded onto the VisionPass devices and enabled for matching only during the period and location of the appointment created by the host. Checks and balances are in place to ensure the system remains secure at all times.”

Products on the market today

As is the norm, we concluded our questions by asking what products the interviewees’ companies have on the market at the moment when it comes to facial biometrics.

Garcia says, “IDEMIA is committed to bringing contactless solutions to the market and we successfully launched our MorphoWave range of products (contactless fingerprint readers) in South Africa over the past few years. We have since grown our contactless range with VisionPass, our facial recognition solution.

“We did not want to bring yet another facial recognition device to the market but really added some key benefits into it. For instance, our reader comes with three cameras (IR, 2D, 3D) which ensure that it can see in most environmental conditions, from very dark to very bright and that it adapts to changes in light conditions within 400 ms. It also ensures our reader has the best performances against spoofing. VisionPass is currently the only facial recognition which is iBeta level 1- and 2-certified against spoofing attacks.

“In addition, VisionPass now uses AI-powered algorithms, which enhances the overall performance of the reader. They use zone of intent and depth of field concepts to ensure VisionPass only identifies consenting users, in the right lane. VisionPass works in different use cases and it is currently in use on estates in South Africa.”

ZKTeco offers a wide range of facial recognition products that can be used in a wide range of applications and environments, states Combrinck. “Our most popular model for residential estates is the ProFace X device that caters for four types of touchless technologies in one device. It has facial recognition, palm recognition, QR code reading and a RFID reader built in. It makes it easy for end users and integrators to adopt these terminals because of the wide range of capabilities on offer in the device.

‍