Video analytics, including face recognition, are at the cutting edge of CCTV, access control, and automated people management systems of all kinds. Their ability to empower existing solutions in extraordinary ways, enhance authentication levels and energise investigations means no new electronic security solution should be installed without analytics coming into serious consideration.

What are the most important aspects of successful face recognition applications and what should integrators and end users be on top of before they even start looking? There’s plenty to think about – not only functionality, but the importance of getting the basics around camera installation right.

According to Gallagher’s Steve Bell, making sure ambient lighting and management of variable lighting in software is adequate is key to accuracy.

“Earlier face recognition solutions suffered from poor lighting of the subject,” Bell says. “Where the light in the environment is relied upon to light the person’s face then there can be variability in performance that meant the sensitivity of the recognition was turned down so that people could gain reliable access. More recent face recognition technologies have solved this problem with smart software and by providing a light source.”

Bell says face recognition and mask detection have helped during the COVID epidemic.

“Managing compliance to corporate policy is often difficult and COVID has created many new risks where the employer has the need to ensure the safety of their workforce,” he says. “A face recognition solution that includes the ability to check for and report the people who are not complying with a mask wearing policy is useful for the employer to help mitigate risk. The addition of temperature sensing is another option that many companies are implementing as an indicator that can be used to help mitigate health and safety risk, where use of this technology is appropriate.”

For integrators, a sales hook with face recognition is its ability to save a customer money by circumventing the need to manage expensive card libraries but there’s cap-ex as well as op-ex to consider. Bell says that currently, reputable face recognition solutions do have a cost that is much higher than traditional solutions for physical access control.

“Many facilities are maintaining cards for a visual indication to all personnel that a person is ‘one of us’ or a visitor,” he explains. “When there is a large number of people in a facility that do not value the visual identifier being worn, then there can be a significant saving in time and materials for managing the card base using face recognition.”

As an access control manufacturer, Gallagher has a strong sense of how interested the end user market is when it comes to face recognition.

“Many markets are interested in contactless biometrics and face recognition has achieved the accuracy and reliability of detection to be one of the simplest non-contact biometric options,” Bell says.

Over at Hills, Orlando Chiang, says important aspects of a successful facial recognition application include the ability to capture, process and match faces within a few a reasonable time.

“In some scenarios when there are too many faces being presented, some systems can’t keep up and this creates a processing queue, which in turn forces the alerts/match events to be delayed,” he explains. “What integrators and end users need to be aware of is how they intend to use the alerts and events generated by the face recognition (FR) system – for instance, how soon after the event do they need to react? This reaction time will determine what solution they go with.

“Not every solution requires real time events. For instance, do they need real time events, or are they content with some delay? If they expect to have high traffic areas with the intention to capture all the faces as they pass in the scene, then they need to expect latency when it comes to matching faces to watchlists, as generally every face presented is compared to every face in the watchlist. This in turn means that they will require higher processing power, compared with a scenario that is capturing individuals being funnelled through designated capture point.”

“Whatever the application, the best face recognition systems must differentiate between a real life individual and a photo of the same individual. They also need to be able to identify the same individual regardless of head cover, face cover or face angle towards the camera. Image quality of incoming video is also essential in achieving high accuracy. This also applies to the enrolling images used to compare all other images. As the saying goes, rubbish in, rubbish out.”

When it comes to the challenges of installing and managing face recognition solutions, Chiang says the initial challenge for any installation is sizing the system correctly for the environment.

“I have seen many systems that have been under sized and performance compromises have been made to satisfy the requirements,” he explains. “I’ve also seen systems that were over-sized with hardware being under-utilised. When it comes to management, the challenge is usually maintaining current usable images that the FR system will use as reference matching images in its watchlists.”

Chiang says it’s important that integrators and end users ensure they get the best of the best in software algorithms when selecting face recognition solutions.

“FR algorithms are constantly being improved,” he says. “Using organisations like NIST and its ‘Face Recognition Vendor Test’ (FRVT) can help narrow choices based on tests. Generally, not all vendors rank high in all tests – even though some vendor algorithms rank high in some tests, it does not mean that the algorithm will work in the scenario specified by the customer – it’s a case of horses for courses.”

There’s enormous flexibility in the application of face recognition analytics – not only for security but in terms of the ability to drive sub systems, report events and streamline investigations. What are the most exciting applications of the technology and which might offer users the best return on investment, in Chiang’s opinion?

“Some exciting applications of FR are in the health care sector,” he says. “FR can be used for tracking dementia patients, alerting staff if they have exited the building on their own, or if they have left with an accompanied authorised person. Another exciting application is in education, where student’s attendance and examinations images are matched with enrolment information, preventing cheating.”

Securing face recognition systems – what’s the key here – how can security integrators and security managers ensure private data stays private – or are privacy fears overblown?

“Face recognition requires the highest level of cyber security,” Chiang says. “Unlike physical access control cards, you can’t just cancel your face. The FR ID that is generated by the algorithm is unique to your face and does not change unless your face changes. Keeping these FR IDs in a secure, encrypted location is essential to prevent identify theft. I have seen demonstrations where these IDs have been captured and used to grant access to hackers in both physical premises and virtual systems. Security of these ID should not be taken as lightly as the protection of passwords have been in the past.”

As the COVID-19 pandemic unfolded, Chiang says it was very interesting to see how quickly FR providers reworked their algorithms to detect the presence or lack of face masks.

“Overall, I do not think there will be much uptake within Australia, since mandatory mask wearing is now relaxed,” he says. “In other countries, which are struggling with the pandemic, the story will be different. In my opinion, some countries are not taking the pandemic seriously, so I would assume their uptake of mask detection may also be low.”

How interested is the end user market when it comes to face recognition – is there hunger for the technology?

“We have seen a lot of interest from the end user market, however, it seems to be an afterthought and a ‘like to have’, instead of designing the whole system around FR to solve a specific operational problem. It’s only once the end user sees the cost savings, that they reconsider and try to find a problem that FR can solve.

“With the current upfront costs of FR, I would say that it would take a long time to pay itself off when compared to physical cards in an access control context, plus the management for both systems will be about the same. What I would say instead is that FR will complement physical cards via dual authentication and help with accompanied access or tailgating, which physical cards can’t address. In scenarios like this FR has paid for itself on day 1.”

Jason Allen of Nirovision says the most important aspects of successful face recognition applications for integrators and end users are the most important aspect of any facial recognition software.

“It starts with how simple it is to build the facial recognition database,” Allen explains. “Can you easily upload names and photos? Can you have people self-enrol? Once people are in the database, how accurate is recognition? Are tools provided to help improve and strengthen identities? Also important is how end users access insights and reports, create and manage watch lists and audit results and if there are integrations available to access control solutions and VMS.

“It’s important to keep these topics in consideration when reviewing the different solutions in the market. Is there a web app and mobile apps so information can be accessed from anywhere and is support available locally to assist if you need help with anything? Hardware is also a consideration. Do you need a server and professionally installed cameras, or can the software run off an iPad that people can present in front of? It’s important to review all these things before considering what other applications a facial recognition system integrates with.

“The best features really depend on what a workplace wants to use facial recognition for. If using it for visitor and access management, the ability to offer different people different workflows – with different processes and questions depending on who’s in front of the camera – is vital to streamline people flow and gather the most appropriate information in each case. Offering users the ability to self-enrol is also desirable, to give them the opportunity to consent to the information being collected and its use. Lastly, having the ability to receive missed registration alerts, sent in the event an individual skips any step of the process, helps audit the process in an unmanned way.

“While most vendors focus on onboarding features, it’s the everyday management that makes or breaks facial recognition solutions. Making physical administration easier with digital processes is especially relevant when working alongside integrations such as access control. For instance, how simple is it to create an identity, and replicate its access levels across systems? How simple is it to enrol and dis-enrol someone?”

What is the greatest challenge of installing and managing a face recognition solution in Allen’s opinion?

“Traditional security cameras are generally not well-placed for facial recognition,” Allen says. “Repositioning cameras or installing new cameras will likely be required if a business wants to track the movement of people around the workplace. Every environment is particular, and changes throughout the day: lightning conditions, types of traffic. These changes, despite being barely noticeable to the human eye, have a tremendous impact on the accuracy of facial recognition results: the more different the faces seen by the cameras from the faces enrolled, the less accurate the results.

“Installations need to be assessed and adjusted once the system is online and detecting faces, especially if there are multiple, disparate cameras in use. Having tools and local support in place to help with the installation, configuration and calibration of cameras and thresholds is crucial to streamline this process.”

When it comes to the underlying software algorithms that deliver face recognition, how can integrators and end users ensure they get the best of the best?

“In the early years of computer vision, accuracy used to be the variable that helped distinguish good from bad performing algorithms,” Allen explains. “A lot of progress has been made since those early days, to a point where hundreds of algorithms surveyed in the latest Ongoing Face Recognition Vendor Test from NIST claim meaningful accuracy figures. With so many options, it’s more important than even to work with a manufacturer committed to innovate, invested in experimenting with the newest technologies and, continuously deploying product enhancements and updates. Working with a trustworthy security integrator is also crucial, understanding how important hardware quality and placement are for the success of facial recognition deployments.

“Image quality is paramount with face recognition, but that doesn’t mean you need the highest resolution cameras to recognise faces accurately. More important is sensor quality – bigger sensors capture more light than smaller sensors – and its relationship to resolution and therefore, pixel size. The better the quality of faces detected, the better the quality of the results. That’s why placement is as important as hardware quality, to ensure faces are being captured front on, sharp and with detailed features. Image focus and blur can have similar effects on performance as very low resolution, so cameras directly looking into sunlight or very dark monitoring areas should be avoided.”

In Allen’s opinion, access control is one of the most exciting applications of face recognition technology.

“This is due to the magic factor of having a door open simply by looking at a CCTV camera or an iPad camera,” he explains. “In more practical terms, this technology also has a convenience factor, as there’s no need to carry yet another swipe card or key, but an organisation is still able to implement enterprise-grade access levels. In addition, face recognition simplifies credential management by reducing the overhead of having to configure access levels individually per door.

“Many businesses, with small and large square-metre footprints alike, do not know the number, nor real-time location of employees and visitors in their premises. Visitor management and time and attendance is another great application for facial recognition, that can save HR and front-of-the-gate teams a lot of time – or replace receptions altogether – by streamlining different instructions and check-in workflows based on whether the person in frame is an employee, a contractor or a visitor. This information can then be leveraged to produce attendance, muster and contact tracing reports, update payroll and HR systems, or inform management of unusual activity in their premises.”

Securing face recognition systems – what’s the key here – how can security integrators and security managers ensure private data stays private?

“The idea of being identified by facial recognition software can make some people feel uneasy, because a face is very personal to an individual,” Allen explains. “Yet facial recognition is very secure and a lot more private than other platforms like social media where we freely upload all sorts of photos and sensitive information.

“Any cloud technology can be vulnerable to attacks but that doesn’t stop us from banking online, or sending important files by email. Why? Because we are confident that security measures that software providers put in place will protect us against hackers and cyberattacks. Facial recognition platforms are similar in that security protocols are followed to encrypt and protect information, but just like online banking and email, there is some responsibility on users to control what security and privacy procedures are implemented to protect information.

“It’s also important to know the origin of the algorithms in use as a lot of products outsource the facial recognition component which means the core technology is not under the control of the enterprise that is selling you. It’s best to enquire if proprietary tech is being used.”

How interested is the end user market when it comes to face recognition – is there growing hunger for the technology in Nirovision’s experience?

“People in general are more aware of the benefits of facial recognition, thanks to smartphones incorporating the technology,” Allen explains. “Anyone using facial recognition to access their smartphone knows the security, convenience and touchless benefits of doing so. According to an Australian industrial workplace survey Nirovision conducted, the majority of respondents (77 per cent) would be very likely or somewhat likely to invest in new technology to help keep their employees and workplace safe, and when prompted on the kind of technologies, (67 per cent) of respondents believe that their workplace would be very likely or somewhat likely to invest in facial recognition technology.”

How useful is face recognition and mask detection during the COVID epidemic?

“Due to its touchless nature, facial recognition is more important than ever in the fight to keep workplaces open and workers safe,” he says. “Whether it’s used for visitor management, worker time and attendance, temperature checks or access control, the benefits are clear and obvious: it can fully automate identification, health and compliance checks, alert upon inconsistencies or problems, and create detailed logs and reports that can be used for contact tracing and auditing purposes.”

According to UNV’s Edward Qui, the most important aspect of successful face recognition applications is accuracy of functionality and the ease of use.

“Accuracy of face recognition systems can reach up to 99 per cent with the advanced algorithm adopted by Uniview and when it comes to ease of use, Uniview’s face recognition access control solutions support indoor and outdoor installation with waterproof shields,” he explains.

“It also includes integration with 3rd party or existing systems, which can save costs through utilization of used equipment. This means integrators and end users should identify their core needs and ultimate requirements first, then find products to meet their operational goals.

“The best face recognition systems should be capable of determining who is allowed to enter or exit, where they are allowed to enter or exit, and when they are allowed to enter or exit – restricting entry to a property, a building, or a room.”

What is the greatest challenge of installing and managing a face recognition solution in Qui’s opinion?

“Actually, the biggest difference between the face recognition access control solution and the traditional IC card solution is that the access controller stores not a series of numbers, but the eigenvalues of a face, which means the data volume of the face access control system is much larger than that of the traditional IC card scheme when large-scale face permissions are required,” Qui said.

“This requires manufacturers to have high requirements on image processing ability. Uniview has many years of accumulation and rich technical reserves in video and image processing technology, which makes our products manage hundreds of face access control, and the face access control system can operate normally.

“When face recognition applied to CCTV systems, image quality plays an increasingly important part. A good image quality captured by Uniview’s access control can greatly increase the accuracy of face recognition and improve the efficiency of retrieval.”

How can integrators and end users ensure they get the best of the best when thinking about FR algorithms?

“Uniview’s algorithm is trained on a large number of faces to ensure that face recognition accuracy is up to 99,” Qui says. “Customers can also test sample before purchase to ensure that the desired results are achieved.”

How can security integrators and security managers ensure private data stays private?

“Users in some countries do have high requirements on the data security of face images, but this, to some extent, is overblown,” Qui says. “On the one hand, few professional security access control systems will be deployed on the WAN, and most will be deployed in the LAN. These systems are generally well protected by firewalls to prevent criminals from invading the security system to obtain user data.

“On the other hand, Uniview’s face access control device uses EMMC storage, which is more secure than SD cards because EMMC can’t be removed artificially. At the same time, Uniview’s face access control can be configured not to store face images, which fundamentally eliminates users’ concerns about the privacy of face images.

How interested is the end user market when it comes to face recognition?

“As biometrics, face recognition is becoming more and more popular. Compared with traditional methods, such as card identification and password authentication, it is more convenient and free from any physical tools – it’s also faster, improving the efficiency of personnel moving in and out,” Qui says. “For these reasons face recognition will be increasingly popular with end users.

“As part of this epidemic prevention and control is a necessary topic nowadays. Face recognition also avoids physical contact between people, and between people and devices, to reduce the risk of infection. Mask detection is also an important measure that needs to be taken during the epidemic period. To remind people to wear masks on specific occasions can play a great role in epidemic prevention.”

BGW Technologies’ Mark Shannon says that as with all technologies, it’s important to ensure that key stake holders have a full understanding of what the end user is wanting to achieve with face recognition, as well as their expectations.

“This way the manufacturer, the distributor, the system integrator and all parties, advise the end user what it will take to achieve those expectations – all parties need to be aligned,” Shannon explains.

When it comes to general features the best face recognition systems should have, Shannon says a solution needs to do the basics right first.

“High accuracy and being able to be integrated into all the key access control and VMS platforms is important, too,” Shannon says. “Just like a card reader is integrated into an access control system or a camera integrated with a VMS platform – it needs to be as seamless as possible and then you build from there. And image quality is critical. Don’t mess with quality if you want accuracy. I cannot put it more simply than that.”

When it comes to the greatest challenges of installing and managing face recognition solutions, Shannon goes back to early expectations.

“The first challenge is to dispel the hype and replace it with reality,” he says. “Secondly, you need to get all parties aligned on what is wanted and what can be achieved. After this, you need to follow up by ensuring the installation meets the facial recognition platform requirements.

“Integrators need an understanding of how facial recognition works in terms of what it needs in order to achieve its peak performance,” says Shannon. “Key, too, is ensuring the system is installed, set up and configured as required. Parameters like the right number of pixels in the field of view for the face, image quality presented to the algorithm and the correct installation angles of the imagers (read that as cameras) form a critical part of the facial recognition system success. We have seen it done right and the outcomes are nothing short of awesome.”

Shannon agrees there’s enormous flexibility in the application of face recognition analytics – not only for security but in terms of the ability to drive sub systems, report events and streamline investigations.

“I think this is where the technology stands out – this is really exciting,” he says. “We have implemented several roll outs that harness the extra capabilities of facial recognition beyond the standard ‘open a door with your face’ scenario. They include finding lost children in shopping centres quickly and easily by scanning an image of child and asking the facial recognition system to scan its history and locate where the child has been and gone. And we are currently rolling out a system where it helps problem gamblers to self-exclude from gaming venues – so much easier than the normal operator being expected to remember a bunch of photos/people. This solution utilises a centralised database, so the problem gambler cannot go to another venue to circumvent their habit. There are so many capabilities right now and there are so many more capabilities coming.”

Securing face recognition systems – what’s the key here?

“Privacy fears are a reasonable concern when you hear about major organisations being breached,” Shannon says. “On each of these occasions, there will always be a situation where something was not updated in time or simply not maintained. However, we only ever hear of the breaches compared to the reality of the many systems not hacked into. So, if the data is encrypted, systems are kept up to date, the latest password algorithms and authentication are used combined with mandated safe practices, this is where concerns will be eased.”

How useful does Shannon think face recognition and mask detection have been during the COVID epidemic?

“Over the last 6 months, a lot of work has gone into this space and detection has now improved – systems and their algorithms have catered for masks well,” he says. “These systems are using AI algorithms, so they do improve from learning. Situations will always pop up where ongoing development will be needed, just like any technology. At BGW Technologies we are lucky – we have an excellent leading Australian face recognition development company that we are able to work with to cater for these situations. The end user market is definitely interested in face recognition technology – there’s absolute hunger – face recognition is hot, hot and hot.”

#securityelectronicsandnetworks.com