Facial recognition is rapidly becoming ubiquitous in our daily lives. From mobile devices to health kiosks, more and more consumers are interacting with facial recognition technology. As the technology becomes more readily available through rapid breakthroughs in AI technology, the use cases continue to expand and you may find yourself exploring your own implementation of a facial recognition solution.
When choosing a facial recognition solution, there are many different factors you have to consider. While each implementation is unique and requires a comprehensive approach, the factors for a successful implementation largely remain unchanged regardless of the implementation. Below, you will find our guide to ensuring a successful implementation of a facial recognition solution.
Facial Recognition is Having Its Moment
In recent years, biometric models, especially facial recognition, have benefited greatly of scientific advancements from Artificial Intelligence (AI), dramatically increasing their accuracy while addressing multiple sources of bias from factors such as skin complexion, age, gender, ethnicity, and even inadequate lighting. Significant improvements on power, cost and size of edge-based hardware have made the technology available to a broadening range of use cases, across multiple industries and for customers of all sizes. With many choices and more options available each month, the question is not: “Can facial recognition be used in your case?” but “What is the best facial recognition solution for your use case?”
Recent Breakthrough and Outlook of Facial Recognition
The recent advances in facial recognition expertise and productization have fueled a fertile innovation cycle across the entire tech industry. Far from slowing down, progress has been accelerating over the past couple of years.
The advancement of both facial recognition software and available hardware have amplified the speed and power at which facial recognition engines can operate today. Now capable of detecting and identifying tens and sometimes hundreds of faces simultaneously, facial recognition has become a critical security asset for large public spaces such as malls, department stores, airports and hospitals, where many individuals may enter simultaneously.
Decreased Cost and Power Requirements
As this technology has continued to be innovated and refined, the costs of purchasing and implementing facial recognition has decreased notably. Thanks to a variety of market offerings with software and hardware bundles tailored to specific use cases, businesses and individuals interested in employing facial recognition have many options available. Similarly, as the technology has advanced, power requirements for operation have lessened, making it available in more settings and with lower recurring costs.
The hardware components needed to run facial recognition are not only seeing dramatic performance improvements but their size keeps shrinking, relaxing an important constraint. Now, facial recognition systems are available in forms as small as an individual electronic door lock.
Public Health Assisted Integrations
As a result of the COVID-19 health crisis, the better facial recognition offerings now come equipped with health tech integrations such as mask detection, facial recognition for individuals wearing masks and social distance monitoring.
Advances that have allowed the technology to become better performing and more affordable have boosted its applicability and allowed for use cases for facial recognition to grow immensely. Where performance has been tested and the value to end-users demonstrated, hesitation quickly dissipates and we see adoption increase . Think of the iPhone Face ID, but also the CLEAR kiosks that are now making their way beyond airport security, into train stations and granting express access to spectator sport and concert venues.
Security and Access Control
As the most accurate & secure biometric technology available on the market, facial recognition has become a critical access security tool, being used to protect secure spaces such as offices, hospitals, banks, warehouses, smart homes and more. Facial recognition provides numerous advantages to legacy access control systems which traditionally use keycards or codes to facilitate entry to restricted areas, by eliminating any possibility of lost, stolen or traded access credentials.
User Experience Enhancements and Personalization
The retail and hospitality industries have become early adopters of facial recognition, leveraging the technology to provide more seamless experiences for both customers and employees. At hotels using facial recognition, guests have the ability to gain access to their rooms and other on-site amenities without plastic keycards, and employees can easily clock in and out simply by entering or exiting facilities through doors equipped with facial recognition cameras. In retail settings, stores can actively monitor for any block-listed individuals or shoplifters while also providing seamless loyalty reward experiences for members without the need for verbally communicating personal information such as phone numbers or email addresses at the register. Similarly, personalized digital signage and marketing opportunities are arising as a result of growing facial recognition adoption.
The Banking, Financial Services and Insurance (BFSI) sector has similarly begun integrating facial recognition into its online banking portals, mobile apps and bank kiosks to provide a high level of security for customers prior to authenticating transactions and other restricted account activities. The high accuracy & security afforded by facial recognition, combined with adjacent technologies that enable the validation of photo ID (e.g., driver’s license or passports) allows any bank or financial service employee to validate their customers’ identity instantly. Furthermore, customers can also validate their identity on their own when using an automated kiosk or even at home through their mobile device. The benefits are obvious, when compared to the traditional tedious checks that typically accompany manual identity validation, such as requesting driver’s license, social security information or confirming numerous security questions which may easily be sidestepped. Even the dual authentication measures often in place today to validate online customers’ identity, such as captchas, tokens or authenticator apps don’t come close to the precision and streamlined experience delivered by facial recognition.
The multiplication of use cases, each with their own sets of needs and constraints, lead to a slightly different question than the one proposed in the opening paragraph: “What is the best facial recognition solution for your unique needs?” As implied in the question, there is not one right answer, but as many answers as there are specific use cases. The remainder of this article will provide a framework aimed at helping you finding YOUR answer.
The Three Facial Recognition Configurations
In its simplest expression, facial recognition typically operates in one of three potential models:
Edge-Based / Single Location
The edge-based, single location model assumes a local server or database (not on the cloud) with which the facial recognition system communicates to verify identification. The edge operation allows for a fast, highly secure process, as the system does not have to communicate with the cloud or any external device. This model is typically used by smaller businesses or, at the other end of the spectrum, to create a highly secure closed-loop system and protect highly valuable assets, for example a bank vault.
Edge-Based / Multi Locations
Facial recognition systems can also operate at the edge while communicating with a central cloud server. This workflow optimizes speed for operators with systems covering multiple locations where a cloud database is necessary.
The cloud-based model requires the facial recognition engine to communicate with the cloud to verify identification. This can incrementally decrease processing speed and security as the system is forced to send data back and forth to the cloud continuously.
An operational facial recognition system consists of the following components:
Edge computing device(s)
A PC, professional workstation, mobile device(s) or IoT connected device(s)
A Central Processing Unit (CPU), Graphics Processing Unit (GPU), Vision Processing Unit (VPU) or a combination.
Edge or cloud.
Peripherals and connected systems
Video/Visitor Management System (VMS), lock control devices, instant messaging, etc.
Facial recognition engine and software
SDK or plug and play software.
The Decision Tree – Navigate the Best Facial Recognition Solution
To build the best facial recognition solution for your unique scenario, you need to consider these key factors:
Use case – what is the primary job you’re implementing facial recognition to do?
Needs and constraints – what specific characteristics does it require? Capacity to recognize people wearing a mask? Instantaneous liveness detection? What are the constraints? Must fit inside a camera casing?
In-house expertise – what is the primary job you’re implementing facial recognition to do?
It’s important to define these factors up-front and create the decision tree that will help navigate the 7 success factors presented below. Addressing the most critical aspects up-front will help discard irrelevant options and focus only on those that matter.
For example, if the use case is a smart lock activated by facial recognition to replace traditional home entrance locks, the form factor constraints (size, shape) will guide focus to small IoT hardware options only, low-cost components, and low power consumption. Meanwhile, considerations around scalability, flexibility and performance of the solution are likely to be less constraining.
A well-defined decision tree also helps address dependencies between the decisions. Here’s how that can work:
For example, if the solution requires a hand-held device with low power consumption and cellular data capabilities, at an initial cost of less than $500, the solution will likely have to run on an Android phone or tablet, limiting each decision to options that fall under these constraints.
The 7 Success Factors for Choosing Your Own Facial Recognition Solutions
There are a several levels to think about when considering accuracy in facial recognition – these include:
Accuracy of the software being run which is impacted by the size of the model, the chipsets and cameras it’s being run on. The FaceMe® model is designed to fit a 4 Mb or 300 Mb database, and is optimized for low power chipsets for maximum flexibility and wide application.
Accuracy of the algorithm itself which is measured by the National Institute for Standards and Technology in their standardized Facial Recognition Vendor Test (FRVT). In the latest FRVT 1:N Identification report, FaceMe® scored an accuracy rate up to 98.11% on its identity recognition against a database of 1.6 million Visa and webcam images, ranking ninth globally out of all tested vendors, and third when excluding vendors from China and Russia.
When Accuracy is a Decisive Factor:
Accuracy is always a critical aspect of the success of a facial recognition system, and for this reason operators are recommended to consider solutions from vendors whose regularly updated algorithms are vetted and ranked highly in industry tests such as the FRVT. Accuracy can be the most critical factor in situations where facial recognition is used to protect access to secure facilities, highly confidential data, controlled or hazardous substances, and other highly restricted tangible or intangible assets. Obviously, the most accurate algorithms, even in cases where they might not be inherently more expensive, require more storage and processing power, which can drive significantly the total cost of deployment as we will discuss later.
Some of the most popular use cases for facial recognition would likely not be economically viable if they used the most precise models available in the market. Other constraints such as form factor, size, weight or energy requirement might force the exclusion of the most accurate models. In practice, most use cases don’t need a 99% or higher level of accuracy. The top vendors in the market likely also offer models that can address all these constraints without losing more than 1 or 2%.
In the below scenarios where accuracy is less dependent, it is still recommended to utilize a solution performing above the 95th percentile.
Highly Accuracy Dependent:
Bank access security – facial recognition is being employed to protect significant financial assets and high accuracy is a dependent factor to successful implementation.
Less Accuracy Dependent:
Stadium turnstiles – a moderate degree of accuracy is necessary so that people don’t have to make several attempts before being accurately identified, but a successful implementation ultimately relies on smooth flow of people, mostly dependent on reliable hardware and the absence of “false positives,” which, in this example would be letting in people who are not registered and authorized as they would be falsely identified as someone who is registered.
Highly Accuracy Dependent:
Large Smart factories – the system is being utilized to protect the machinery and personnel within - liability is high. The success of the facial recognition system is hinged on high accuracy.
Less Accuracy Dependent:
Smart homes – the smaller quantity of visitors and less incentive for bad actors makes this application less dependent on maximizing accuracy. Cost, form factor and ease of implementation are likely to be prioritized, for example in the case of smart locks.
Size of the deployment:
Highly Accuracy Dependent:
Large department store – potentially comparing against a national database of VIP/loyalty customers or block-listed individuals, the success of the system depends on highly accurate identification across its many stores.
Less Accuracy Dependent:
Local shop – with less customers and likely a smaller database to identify against, accuracy is less of a constraining factor.
Facial template extraction
Face compare & match
Anti-spoofing with 2D camera
Anti-spoofing with 3D depth camera
Face with Masks
Facial recognition through masks – TAR up to 98.21%
Each facial recognition solution offers specific features . The three basic features that should be expected from any solution include:
Face Detection Face detection is the first step the technology takes to identify a face. In this step, the technology works by scanning the whole image to see if any area contains full or even partial human faces. Fast and precise face detection is a critical first step to ensure the performance of the entire facial recognition process. Leading solutions, like CyberLink’s FaceMe®, can detect multiple faces at once, count how many faces are present and perform detection on each of them individually.
Face Recognition After detecting the face, the software looks at unique information around specific facial features. This information template is then matched to those pre-enrolled in a database to match to the correct person’s identity, if included. Given the controversy that sometimes ties facial recognition to personal privacy, we strongly advise that you select a vendor using a high standard of encryption making the template data unusable to anyone not authorized to use it. When using a highly encrypted template, no actual face images need to be stored through the platform, ensuring full privacy protection.
Face Attribute Detection Face attribute detection is the task of identifying and analyzing characteristics such as age, gender, facial expression and head orientation or movements (e.g., nodding, shaking). This feature is a key enabler of smart retail and digital signage, for use cases such as pushing dynamically customized ads and messaging to micro targeted audiences, or collecting detailed visitor statistics.
The most advanced facial recognition solutions such as FaceMe®, also include enhanced features, such as:
Image Enhancement This process enhances the quality of images captured by poor lighting or camera quality and enables a high degree of precision when processing this information for facial recognition.
Anti-Spoofing Anti-spoofing can provide liveness detection using either 2D or 3D cameras. When using a 2D camera such as a USB webcam, spoofers are caught through interactive and non-interactive measures. Interactive measures detect natural and precise head or facial movements to confirm the presence of a live person. Non-interactive measures are unique to each solution provider and their AI algorithm for face detection and recognition.
3D cameras perform depth detection, allowing quasi-instantaneous anti-spoofing. There is no need for interactive detection or recognition measures. 3D cameras generally provide a superior experience but are costlier, while 2D alternatives can also provide accurate anti-spoofing at a fraction of the cost.
Mask Detection and Facial Recognition with Masks Mask detection features designed for public safety and health application can detect the presence of a mask and verify whether individuals are wearing the mask correctly over their nose and mouth. Some advanced solutions, such as FaceMe®, also enable high accuracy facial recognition while masks are worn.
When Features are a Decisive Factor:
Highly Feature Dependent: Access control for a secure warehouse – anti-spoofing would be an important feature to ensure the system cannot be bypassed by photos/videos of approved personnel.
Less Feature Dependent: Recognition for retail loyalty programs – anti-spoofing is likely to be less important in this scenario as it’s unlikely individuals would attempt fraudulent verification.
Highly Feature Dependent: Smart city – the mask feature may be critical for a smart city, public health and safety deployment, especially in our current environment.
Less Feature Dependent: Smart home – for a smart home, mask detection is not necessary since individuals do not remain masked in their private homes.
Highly Feature Dependent: Shopping mall – having the feature ability to scan multiple faces concurrently may be critical for a shopping mall deployment when large groups of people are being scanned for potential block-listed individuals.
Less Feature Dependent: Individual retail store employee entrance – In a standalone retail store using a facial recognition system at a single door employee entrance for identity verification and clock in-clock out, it’s likely only necessary to scan one individual at a time as they enter and the feature of scanning many faces simultaneously may not be needed.
As with accuracy, many components play into the performance ability of a facial recognition system. Let’s break them down here:
Frames per Second/ FPS This is a measurement of the amount of pictures being taken per second by the cameras, and then being communication to the facial recognition system for processing. Higher FPS can provide higher accuracy and performance.
Detection Speed The detection speed measures how quickly the system can scan and detect facial features in a space and recognize that there are faces present.
Extraction Speed Extraction speed is the time the facial recognition system takes to then extract the information about the face that it will process for identification.
Recognition Speed Finally, recognition speed is the final part of the process, and measures the speed at which the system can then process the extracted information and deliver either an identification or no-match.
Why Performance Matters:
Performance can be a dependent factor to enabling a facial recognition system in multiple use cases. For example, deployments in large facilities often need to handle hundreds of video channels concurrently, some in high traffic areas. A vendor offering high performing facial recognition models can significantly reduce the number of expensive workstations required to monitor such facilities.
Performance and Edge vs Cloud Architecture:
In the next decision section, we’ll be looking at edge vs cloud architecture for your facial recognition system. It’s important to consider that edge systems, while not having a inherent performance advantage over cloud-based processing units, generally deliver much faster facial recognition as sending images or video to be processed on the cloud increases response time from milliseconds to several seconds.
Benchmarks for Best Performance:
Performance can be enhanced depending on chipset selection. Using standalone GPU or VPU chips like the NVIDIA T4 with a separate CPU can boost performance.
There are multiple option for GPU acceleration. Harnessing OpenVINO, NVIDIA CUDA/TensorRT, Intel Movidius, NVIDIA Jetson, Qualcomm SNPE, MediaTek NeuroPilot and more can speed up deep learning algorithms and further optimize performance.
Facial recognition software options are each unique and have varying levels of optimization with different chipsets and system architecture. For example, FaceMe® has optimized system architecture through several iterations to ensure it delivers the absolute best performance. On a single workstation, FaceMe® with NVIDIA RTX A6000 can handle 340-410 frames per second (with the exact number depending on which FaceMe® facial recognition model is used). This is equivalent to handling 25-41 concurrent video channels (each with 10 frames per second) per workstation – a high performance option.
When Performance is a Decisive Factor:
Highly Performance Dependent: Airport deployment – in a crowded public setting like an airport, a facial recognition system is employed to scan tens and sometimes hundreds of faces concurrently in the space to watch for blocked individuals or mask wearing. The huge volume of data that needs to be processed simultaneously could easily require a cost-prohibitive dedicated hardware deployment in absence of the significant performance advantage delivered by the best facial recognition solutions.
Less Performance Dependent: Library deployment – in a theoretical use case where facial recognition is being deployed in a library setting, scanning faces individually as they check books out, performance is less of a constraint as the system is processing less information at once.
Highly Performance Dependent: Warehouse/logistics – with many individuals present in a large facility at once and multiple camera feeds directing to the facial recognition system, performance is an important factor.
Less Performance Dependent: Small office access control – facial recognition systems providing access control in small offices and for individual door entry generally have less performance constraints as they are likely only processing information about one or a few faces at a time.
Highly Performance Dependent: Large facility with multiple video feeds – additional video feeds put pressure on processing time and performance for the facial recognition system. To operate successfully, higher performance chipsets and software should be considered.
Less Performance Dependent: Small facility with one video feed at entrance – single video feed systems do not put pressure on system performance and in this case, performance is not likely to be a top dependent factor when selecting system components.
4. Edge vs Cloud:
Edge vs cloud architecture can impact the security and performance of your facial recognition system as a whole and can be an important consideration for operators seeking maximum speed. Edge-based systems operate more quickly because information does not have to be sent back and forth to the cloud, potentially adding several seconds for upload transmission.
Edge-based systems offer several additional benefits and thus have become popular choices:
Security – edge-based systems don’t have to send vulnerable information to the cloud where it could be intercepted.
Flexibility – edge-based systems don’t need to be connected to the cloud, making them more flexibility installed in a variety of settings and use cases where cloud access may not be available.
However, cloud can be a better option to consider in use cases with:
Infrequent use – such as protecting a facility not frequently visited.
Tolerance for lower accuracy – in lower risk deployments such as retail loyalty programs.
Significant hardware cost constraint – in cases where your existing hardware cannot be replaced and depends on cloud infrastructure.
5. Device and Hardware Support
The hardware environment is sometimes a constraining factor when selecting your facial recognition system. Here’s what to take into account:
Operating System Support
Windows and Android are the most common operating systems supporting by facial recognition software. Better solutions offer a wider range of OS support such as iOS, Linux versions, Jetson, etc. For example, FaceMe® offers one of the market's most comprehensive support, with more than 10 OSs: Windows, Android, iOS, Linux variants, Ubuntu x64, Ubuntu ARM, RedHat, JetPack (mainly for NVIDIA Jetson family), CentOS, Yocto ARM.
Due to their reduced size and capabilities, iOS and Android have greater restraints in terms of processing power as compared to Windows or Linux which are much more powerful.
Large facilities, for example petro-chemical plants which house a variety of security systems and existing IoT infrastructure may be running multiple OS across these systems. In this case, it’s important to consider support for interoperability between multiple operating systems.
When Device & Operating System Support is a Decisive Factor:
Highly Dependent on Device & Operating System Support: Large bank – in this use case, with a large facility likely running multiple video feeds and requiring high accuracy to protect the financial liability within, hardware considerations for higher performing OS are necessary. Windows or Linux options are likely to be a better fit here as compared to mobile iOS or Android.
Less Dependent on Device & Operating System Support: Smart home – in a smart home deployment for personal use, there are much lower performance and accuracy requirements. In this case, hardware cost, size and convenience are likely more important deciding factors.
Highly Dependent on Device & Operating System Support: Smart college campus – a facial recognition system performing across an entire college campus will likely require high performance, moderate to high accuracy and be capable of supporting many OS’s, to integrate with other legacy systems running on the campus.
Less Dependent on Device & Operating System Support: Smart office – in a single office integrating a facial recognition system for the first time without any existing systems and without extreme accuracy or performance requirements, hardware is less of a constraining factor.
Highly Dependent on Device & Operating System Support: Legacy hospital environment – in this setting with multiple buildings and security systems for each running on different operating systems, interoperability of OSs will be an important consideration for implementation.
Less Dependent on Device & Operating System Support: Single doctor’s office – a single building with one system running on one OS has more hardware flexibility.
Thanks to fast evolving hardware and chipset technology innovation, there are ever increasing options on the market to best address speed, power, form factor and cost constraints, driving the introduction of new use cases previously not possible.
Hardware Options to Run Facial Recognition
PCs are most commonly used for facial recognition by smaller operations or for single use cases. Take a store or restaurant that wants to identify VIPs, automatically clock-in employees or get alerts for block-listed individuals.
Organizations looking to deploy facial recognition for security monitoring and access control over tens or hundreds of video channels across large facilities will benefit from one or several workstations powered by a high-end GPU capable of handling multiple IP camera video feeds simultaneously.
Servers are useful for situations where there are many video streams and photos or video coming from independent devices such as mobile phones, and which need to be processed quickly and powerfully while still using the cloud.
Kiosks & Smart AIoT Devices:
The fast-paced edge-computing innovation that drives performance while cutting costs, is opening the door to endless IoT device use cases powered by facial recognition such as smart kiosks. For example, frequent travelers are all too familiar with Global Entry and Clear kiosks, both using facial recognition. Now smart kiosks integrating facial recognition are being deployed in fast food restaurants, hospitals and hotels. AIoT devices may house a small computer, for example a NVIDIA Jetson or an Android board, or integrate the processing and storage units into the device’s electronic board.
Chipset Options to Run Facial Recognition
CPUs are typically a simpler, cheaper option.
Facial recognition systems can also operate at the edge while communicating with a central cloud server. This workflow optimizes speed for operators with systems covering multiple locations where a cloud database is necessary.GPUs are typically more expensive and higher performing.
GPUs are typically more expensive and higher performing.
Combinations of different chipsets are designed to power complete yet affordable solutions.
When Hardware Support is a Decisive Factor:
Highly Hardware Dependent: College campus health and security desk – a central desk monitoring security and potentially mask usage in a pandemic scenario across an entire college campus needs to incorporate hundreds of cameras and is likely to be dependent on powerful hardware. A workstation or server may be good options to consider in this use case.
Less Hardware Dependent: Apartment building smart locks – hardware selection for facial recognition-powered individual dorm door locks would be more dependent on size as and have less constraint on performance.
Highly Hardware Dependent: Hospital facial recognition system for security, access control and health monitoring – Running many video feeds checking for identity and masks simultaneously, this vertical application is likely to require more robust hardware such as a workstation.
Less Hardware Dependent: Individual retail store – this smaller application scenario running less video feeds and with less pressure on performance is likely to have a greater focus on cost and convenience when selecting the appropriate hardware. A PC is likely a convenient option for this vertical application.
Highly Hardware Dependent: Monitoring a nationawide chain of large retail stores – taking in hundrers of video streams and photos from IP cameras in each store, this large application scenario is likely dependent on a hybrid model that includes high-performing workstations in each store to perform the face detection and extraction, combined with centrally-located high-powered servers to match the captured facial templates sent from each location with a central database.
Less Hardware Dependent: Monitoring an individual hotel – on a smaller scale, implementing a facial recognition security system into an individual hotel puts less pressure on performance and may be appropriate for a high-powered PC or a workstation.
6. Software and SDK Flexibility
Facial recognition software is the actual program which processes information extracted from video feeds to determine a match or detect faces. There are a few formats in which the software component of your facial recognition system can be incorporated – let’s walk through them further:
Plug and Play Software For a while, facial recognition solutions mainly came in the form of a software development kit (SDK). They are generally flexible and allow for the deployment of perfectly tailored solutions, but require significant programming and integration work. This is why plug and play software can be a great option for well-defined use cases and quicker timelines. Options like FaceMe® Security, a ready to go, plug and play facial recognition security software, are pre-set to best accommodate typical security use cases like access control and monitoring. Plug and play solutions have the internal software infrastructure already set up for easy implementation. The better offerings are highly scalable and can be used in a single camera scenario, all the way to multi-camera, multi-location deployments. They can typically connect into existing cameras and networks. The very best solutions can connect easily into other systems such as VMS, door locks, time & attendance software, etc.
Software Development Kits (SDKs) SDKs are highly flexible and the de-facto option for unique scenarios where you want complete control to mold the facial recognition algorithm to your software infrastructure. It’s important to note that to effectively implement an SDK, you will need more robust internal computing or IT talent to integrate the SDK into your existing software infrastructure. SDKs allow organizations incorporating facial recognition to leverage it with their existing workflows and processes.
When Software Format is a Decisive Factor:
Highly Software Dependent: Integrating FRT for patient management and access control in a hospital environment – in this use case where a facility depends on a series of uniquely designed processes and systems, often each running on their own platform, an SDK would be a more flexible software format.
Less Software Dependent: Adding FRT to a retail store or a chain of standardized stores for security and access control – whether there is an existing camera security system or a video management system (VMS) connecting the cameras, a plug and play solution like FaceMe® Security will likely offer an attractive solution that requires minimal deployment lead time, is cost effective and needs little to no maintenance.
Highly Software Dependent: Access control at a bank – if a retail bank wants to change its entrance readers from a credit card scanner to facial recognition, they’d likely need to incorporate it with an existing systems and enterprise infrastructure. In this example, an SDK would likely be a better fit.
Less Software Dependent: Security and access control in an office building – Typically office facilities have relatively similar needs around security monitoring, access control for employees and visitors and alerts for block-listed people. Larger facilities generally have a VMS with security cameras and door access control already, making it easy to connect with a leading a plug and play software option like FaceMe® Security. And for facilities that don’t have a VMS, FaceMe® Security was recently updated with it Monitoring Add-On that emulates the key monitoring and alert displays from VMS.
In many cases size is not a barrier to using a plug-and-play solution, as long as you are looking to address a relatively standard set of functionalities or use cases. Good software solutions are scalable. Unique customized solutions integrating an SDK generally come with significant costs of integration and are typically reserved to large organizations. However, with the fast decrease of hardware cost and the exponential improvements in chipset performance, Facial recognition SDKs can now be integrated at a reasonable cost in mass-market AIoT products that can sometimes retail below $100.
7. Cost of Ownership
It’s important to consider cost of ownership for the full life of your facial recognition system prior to integration. Here are some of the core components to consider:
Initial Costs Initial costs include the one-time expenses and investments made when you first begin to design and integrate your facial recognition system. This may include but not be limited to: research, PoC, Hardware, software, integration, training, initial data creation and legacy equipment retrofitting.
Recurring (Variable) Costs Recurring costs will continue through the life of your facial recognition. They may include system maintenance, the cost of subscription to the facial recognition software, monthly bandwidth and energy expenses, server rental and cost of the capital involved.
Obsolescence It’s necessary to make timely upgrades to equipment, operating systems and software throughout the life of your facial recognition system to avoid obsolescence as a result of outdated components
Replacement Cycles Replacement cycles are a popular maintenance option to benefit from lower cost over the long run. By systematically replacing components of your hardware and software, you can ensure you are using the most up to date technology which has been optimized for better performance and lower costs (including costs of upkeep, less energy usage, etc.).
Replacement Cycles Many cost drivers are tied to the size of your deployment scenario and should be taken into account when considering overall cost estimations. For example, the more buildings you are securing, the more hardware stations required, higher cost of facial recognition software and higher monthly costs for maintenance, energy, bandwidth, etc.
Here are some examples of cost relative to deployment size:
Small shop, 1 location:
This scenario would have low costs relative to size, including:
Software: Plug-and-play software pricing often tied to number of video feeds is low when there are only a few cameras
Hardware: Less expensive hardware needed. Often a reasonably performing PC will do.
Training: Low integration and training costs if using plug and play software, often included in the packaged offered by the VAR that sold the system
Monthly costs: Minimal monthly energy
Chain of small shops, multi locations:
Costs increase incrementally from our previous scenario:
Software: Higher cost driven by more video feeds
Hardware: Typically a PC or a low-cost specialized computer (e.g., NVIDIA Jetson) and a few cameras at each location, with a similar setup at each location. At least a server at one location if there is shared customer, employee or block-listed people data.
Training: Reasonable training and integration costs from the VAR, driven by the number of areas/shops and the geography covered
Monthly costs: Relatively low monthly energy and bandwidth used, largely driven by the number of locations
Large facility, 1 location (e.g., a factory)
Incrementally more than our last option:
Software: Relatively low monthly energy and bandwidth used, largely driven by the number of locations
Hardware: Typically one or more workstations with multiple GPUs or VPUs will be needed, together with a large camera deployment at that one facility
Training: Moderate to high training and integration costs to learn a more complex system
Monthly costs: Monthly energy costs. Potentially monthly maintenance contract with integrator or VAR
Large facility, multi locations (e.g., national grocery store chain)
Most expensive scenario:
Software: Top tier software or SDK, often billed in function of the number of video feeds, with minimal (software) to potentially significant (SDK) installation cost with system integrator, tied to the number of locations and the size of each location’s deployment. The software or SDK pricing will typically grow with the number of locations, although customers will generally benefit from volume discounts. Deployment can be significant, although expected and consistent with costs associated to other deployments
Hardware: Typically one or more workstations with multiple GPUs or VPUs will be needed, together with a large camera deployment at each location. In addition, each location or regional center will probably need one or more servers for database hosting and sharing
Monthly costs: Significant bandwidth and energy usage. Likely monthly maintenance contract with integrator
Designing the Best Facial Recognition System for You
There are many options available when it comes to designing a facial recognition solution for your unique scenario. Each core component and decision element described above will impact the effectiveness of the solution for your needs. A safe approach starts by familiarizing yourself with each decision and the range of options that exist today, consider how fast the technology and solution offerings have improved in the recent past and inquire about trends and how expected breakthrough might inform your testing and deployment timeline. Then focus on the elements that are more crucial for the success of your deployment (such as performance, features, hardware, etc.).
It can also be useful to learn about successful deployments and failures in your industry. Many industries have associations and task forces monitoring and analyzing facial recognition and how it could benefit their members.
Finally, use the culmination of this information to build your unique blueprint or decision tree that will guide the right decisions and ultimately result in the best solution that fits your unique needs.