article


Article: Intelligence and Efficiency through On-Demand Media Analysis using Face Recognition

You can download a PDF copy of this article by clicking this link.

Governments are implementing the most severe budget cuts of recent times. Against this backdrop, threats from terrorism, organised crime and public disorder continue to rise. Yet recent statistics in the UK demonstrate that authorities can remain resilient and still ensure law-and-order. The targeted application of technology can further increase resilience and the readiness to respond to major events. The relentless advance in the accuracy of face recognition technology, increase in the availability of digital media and mass availability of cheap computing power now provide unique opportunities to meet challenging budgets by drastically enhancing the operational efficiency of forensic investigators while even further enhancing public safety. Digital media can be bulk-ingested in an automated fashion to be processed in a cloud computing environment to identify and extract potential actionable intelligence. Processing is continuous, consistent and predictable. Multiple identification technologies can be deployed and the most suitable algorithms integrated to meeting evolving requirements. Analysts can now focus on investigating and confirming suggested results rather than having to manually watch countless hours of media in the hope of stumbling across the required information. Expanding beyond traditional sources of media is increasingly being accomplished by engaging the public and crowd-sourcing intelligence in response to incidents.

Having previously written on the subject of the application of face recognition in airports[i] and privacy concerns of face recognition when used by retail[ii], this article focusses on the application of face recognition to support bulk processing of media by what has traditionally been the first and thus far most proliferate user of biometric technologies: law enforcement. The convergence of multiple advancements now provides a whole new set of opportunities to use identification technologies in manners that provide benefits that are only now being realised.

1.A Need for Enhanced Safety and Operational Efficiency

Austerity is the order of the day and public budgets are being slashed. Against this backdrop, security risks are continuously increasing. The threat from terrorism, organised crime and public disorder is not abating. Indeed, as reported by the BBC News on the 17th July 2013[iii], the threat landscape is “substantial” and becoming ever more fragmented, consisting of a greater number of smaller and less sophisticated plots.

However, the UK’s police forces have demonstrated that it is possible to maintain and even improve upon public safety despite the relentless pressure of austerity. Recent reports indicate that crime in the UK is at an historic low, being at its lowest level since 1981 [iv]. As always, it appears that necessity is the mother of invention and it is likely that technology is playing an innovative role in improving police efficiency.

What is not apparent from these recent reports, however, is the current level of readiness to respond to a major event. Indeed, the UK’s Police Federation, the body representing rank and file police officers, warns that the police “could not handle more riots”[v] after the budget cuts and Her Majesty’s Inspectorate of Constabulary (HMIC) warns that neighbourhood policing risks being “eroded”. [vi]

There is a need to enhance public safety whilst reducing public operational budgets.

2. A Relentless Increase in Digital Media

The increase in the creation of digital media is relentless. Law enforcement and intelligence agencies have amassed large collections of biographical, video and photographic information from multiple sources such as:

  • Computer hard drives.
  • Mobile phones and portable cameras.eye
  • Flash memory devices.
  • Online sources on the Internet such as Facebook and YouTube.

Additionally, when tragic events or social disorder occur, investigators have a long and arduous task of reviewing countless hours of CCTV footage, generally with a varying degree of concentration and scrutiny.

A solution that minimises manual effort in the extraction of actionable intelligence from amassed media by automating this process with a consistent and repeatable level of scrutiny will deliver concise and consistent information in a fraction of the time taken by operators undertaking the task completely manually.

3. An Automated Media Processing and Exploitation Solution

Police, intelligence and other public order agencies would benefit from the application of a powerful media processing solution designed to process, ingest, analyse and index in an automated fashion very large quantities of photographs and videos to transform them into usable assets. 

Allevate is offering MXSERVER[vii], already operationally proven with US Federal agencies, to EU agencies to address the issues raised in this article..

Such an automated solution ingests and processes media from multiple sources. Once processed, law enforcement agencies can analyse and make use of the extracted assets and manage them in a centralised repository of information. Data links, associations and metadata inferences can be managed across the whole dataset by multiple users from a single common user interface. Backend processing services are run in a cloud-computing environment, the capacity of which can be configured and incrementally scaled up and down to meet an organisation’s changing demands; peaks arising from specific events can be easily accommodated.

Features include:

  • Automatically find, extract and index faces to enable biometric and biographic searching of media.
  • Create and manage watchlists of people of interest via a web-based interface.
  • Find and cross-reference all media instances in which a person of interest has been seen.
  • Identify, locate, and track persons of interest, their associates and their activities across all media.
  • Discover, document and view links between people of interest, their activities and networks.
  • Use of metadata (including geo data) in the media to enhance investigations and association of data.
  • Integration into existing system environments, databases and components via a flexible API.

 

3.1  Incorporating Other Detection Capabilities

In addition to face detection and recognition, other detection engines can be incorporated, such as:

  • Automatic Number Plate Recognition. (ANPR)
  • Voice Biometrics.
  • Object / Logo Recognition.
    (Other identifying features can be used to track individuals through other processed media.)
  • Scene Recognition
    (Identify similarities in the entire frame, often used in child-exploitation investigations)

Vendor independence allows the use best-of-breed algorithms.

3.2  Biographic Filtering and Fuzzy Match Capability

Forensic investigations are complex and require a holistic view of all available data. This involves not only analysing media, but making full use of all textual and biographic data available as well. This can include text from files recovered from hard drives and other storage devices, online sources, metadata associated with photo or video files and data entered by investigators during the investigation.

Traditional Boolean search techniques only work within a black and white, true and false paradigm. More applicable within a complicated forensic analysis are techniques that use advanced “fuzzy” algorithms that to calculate similarities and aggregate match scores using multiple criteria to enable a “shades of grey” analysis.

Such an approach can fuse match scores across multiple disparate search criteria and even allows for fusion and aggregation of search results across multiple biometric and biographic criteria.

The use of media metadata and other biographic data further refines biometric matching.

3.3  Working with Geo-Location Data

An ever-increasing amount of media available to investigators is captured on mobile devices and cameras affixed with location determining technology. This includes media obtained from CCTV, confiscated hardware and devices, online sources and voluntarily made available by members of the public. The majority of the time, this geo-location data is incorporated into the media metadata, thereby providing significant potential to further enhance the analysis of the media. For example, geo-location can be used to:

  • Compartmentalise and refine analysis by location of where the media was created.
  • Overlay location of proposed matches onto maps.
  • Chart movements of individuals of interest by location and time of sightings.
  • Link individuals at the same location and time even if they do not appear together in media

3.4  Architecture and Integration with Existing Systems

There are significant similarities in organisation and methods of operation in many western law enforcement agencies facilitating increased levels of co-operation. Operational systems should support full control of information and data as well as have sufficient in-built flexibility to enable authorised data exchanges.

In addition to utilising COTS components, adhering to common standards and being cloud-architected to enable massive scalability, a well delineated scope of functionality and open API enables:

  • Flexibility in customisation and integration with existing systems and workflows.
  • Well-defined mechanisms of loading data and automating ingestion of media for processing.
  • Dynamic alteration and sharing of watchlists, media, system-generated results and operator analysis.

3.5  Hosting, Cloud and Virtualisation Options

Full architectural flexibility enables flexibility of hosting options. Organisations can elect to:

  • Take advantage of IaaS and SaaS options on cloud offerings.
    (UK accreditation of IL0 to IL3 is available via hosting partners)
  • Fully host the solution on their own private and secure premises and datacentres.
  • Deploy in a hybrid manner.
    (Thereby taking advantage of external processing power whilst retaining the most secret data) 

3.6  Working Hand-in-Glove with Trained Forensic Investigators

The human operator will always remain the critical and essential part of intelligence analysis; media analysis solutions are not designed to replace the intricate skills and knowledge of trained investigators. Rather, the operator is enabled to intelligently direct and apply their extensive training at suggested results, eliminating the necessity of rote viewing of countless hours of media either in a sequential our random fashion.

Integration of enhanced verification, charting and mapping tools enables operators to conduct detailed analysis of suggested matches and identifications to confirm or deny them.

4. Potential Use Cases

There are myriad different applications of a solution architecture as described herein within military, law enforcement, intelligence and public site security agencies. These are summarised into three broad categories:

4.1  Time Critical Investigations, Media of Critical Importance

In certain major incidents, timeliness of response is of the essence. Authorities need to quickly process evidence to identify and apprehend individuals. The scale of the investigation is often huge and the amount of media that needs to be processed massive. Examples include terrorist events such as the recent Boston bombing and the Woolwich attack in South London.

Often, the media acquired in these instances is of such critical importance that the authorities may choose to review it all in its entirety, frame-by-frame. However, in the early stages after the incident, decisive and immediate action is critical. Rather than having to sift through the media in a random or sequential fashion, a media analysis solution can quickly direct the investigators to the portions of the media that are most likely to deliver immediate results. Full review of the media can be conducted during subsequent stages of the investigation.

4.2  Bulk Ingestion of Media Arising from Criminal Investigations

During routine operations or specific criminal investigations, authorities may recover significant quantities of media on confiscated hard drives, mobile phones, flash / thumb drives and other sources that need to be processed to either further the investigation or to assist in building an evidence base for criminal prosecution.

Examples include:

  • Military or counter-terror officers raiding terrorist training facilities.
  • Specialist organised crime investigators raiding the offices of organised crime syndicates.
  • Child protection officers raiding premises of individuals or organisations involved in child exploitation.

This media can be bulk ingested in an automated fashion to provide the investigating officers an overall summary of the contents including focus areas for further investigation.

4.3  Continuous Background Processing of Media Sources

Authorities may as a matter of routine have access to masses of media which may contain actionable intelligence, but typically would never be viewed or processed due to a lack of resource and the time consuming nature doing so. Examples include:

  • Media from specific cameras installed at high-profile or sensitive locations.
  • Media from known or suspect online sources or accounts from social media sites.
  • Media made available to the authorities by the general public.

Intelligence in these sources may be missed entirely and never acted upon.

This media can now be bulk ingested and processed in an entirely automated fashion to flag any relevant intelligence, using operator controlled criteria, to the authorities as required for follow-up processing.

5. A Compelling Business Case

The solution and optional IaaS / SaaS components can be made available on a monthly service-charge basis, thereby requiring a minimal capital outlay and enabling a compelling operating expenditure business model.

Whilst the human operator is an essential part of intelligence analysis, an entry level solution empowers the analyst to process up to an order of magnitude more media on a daily basis. This enables trained operators to apply their expertise and training by focussing on the analysis of results generated by the solution in a more focused effort than manually watching hour upon hour of media.

Efficiency is dramatically boosted by bulk processing media 24×7 at a constant and predictable level of focus and accuracy: operational staff can focus on analysing results.

6. Engaging the Public to Crowd-Source Media to aid Investigations

Increasingly, especially from crowded public events, authorities are making greater use of media captured intelligence in the form of photographs and videos that have been recorded by members of the public.

With the advent of smartphones, almost everybody has a high quality camera in their pocket.

Most members of society would welcome the opportunity to assist the authorities with their investigations, but often do not know how or are fearful of being involved.

Allevate’s proposed PublicEye service is aimed at empowering the public to take a greater collective social responsibility and assist law enforcement in much the same manner as the phenomenally successful CrimeWatch. It enables members of the public to (at their discretion) upload media directly from their mobile phone or other internet device to a public portal for processing and dissemination to the relevant authorities.

A PublicEye portal could be used:

  • In response to appeals by the police to the public who were present at an event or disturbance.
  • When individuals witness a crime being committed.
  • Upon suspected sightings of missing persons or individuals wanted by the authorities.

A PublicEye enables the authorities to crowd-source media to augment their own sources.
 

7. Summary

In the UK, MXSERVER is available on the G-Cloud catalogue [viii]which is designed to provide a streamlined process for buying ICT products and services as a commodity without having to invite tenders from suppliers.

Security concerns are ever increasing. However, public budgets are being slashed. Law enforcement agencies are rising to the challenge of implementing budget cuts partly through the focussed application of technology. The accuracy of face recognition has increased dramatically over the past 10 years. This, coupled with the massive increase in the creation of digital media and the availability of cheap computing, now provides authorities with the ability to bulk ingest and process media in an automated fashion. Results are continuous and predictable. Trained analysts can now focus their skills on investigating suggested results and on intelligence extracted by automated systems. Not only does this provide the ability to process critical media even faster than ever before to respond time critical investigations, but it also enables authorities to extract intelligence from media sources that in the past may never even have been looked at because of the significant resource this previously would have entailed.

Additionally, the availability of smartphones means almost everybody is carrying in their pocket a high quality camera. The ability to process media rapidly and cheaply means the authorities will be able to, on a continuously increasing basis, engage with members of the public to crowd-source media in response to major investigations.

8. About the Author

Carl is the founder of Allevate Limited (http://allevate.com), an organisation that works with law-enforcement, intelligence and government agencies to enhance public safety by ensuring positive identification through the application of biometric and identification technology. With over 20 years’ experience working in the hi-technology and software industry globally, he has significant experience with identification and public safety technologies including databases, PKI and smartcards, and has spent the past 10 years enabling the deployment of biometric technologies to infrastructure projects. Carl started working with biometrics whilst employed by NEC in the UK and Allevate subsequently supported NEC’s global and public safety business internationally.

Residing in the UK since 1993, Carl was born and raised in Canada and holds a Bachelor of Science Degree on Computer Science and Mathematics from the University of Toronto.

You can download a PDF copy of this article by clicking this link.

 Follow us on Twitter: @Allevate

2,774 words


 

[i] Allevate, July 2012:

[iii] BBC News, 17th July 2013
http://www.bbc.co.uk/news/uk-23334719

 

[viii] The UK’s G-Cloud Programme is a cross government initiative led by Andy Nelson (Ministry of Justice) supported by Denise McDonagh (Home Office) under the direction of the Chief Information Officer Delivery Board as part of the Government ICT Strategy.

http://allevate.com/news/index.php/2013/05/13/face-recognition-media-exploitation-system-g-cloud-iii-cloudstore/

 


Article: Face Recognition: Profit, Ethics and Privacy 2

You can download a PDF copy of this article by clicking this link.

The accuracy of face recognition has increased dramatically. Though biometric technologies have typically been deployed by governments and law enforcement agencies to ensure public, transport and border safety, this improvement in accuracy has not gone unnoticed by retailers and other commercial organisations. Niche biometric companies are being snapped up by internet and social media behemoths to further their commercial interests, and retailers and other enterprises are experimenting with the technology to categorise customers, analyse trends and identify VIPs and repeat spenders. Whilst the benefits to business are clear and seductively tantalising, it has been impossible to ignore the increasing murmurs of discontent amongst the wider population. Concerns over intrusion of privacy and the constant monitoring of our daily lives threaten to tarnish the reputation of an industry which has endeavoured to deliver significant benefit to society through improved public safety. Can the industry be relied upon to self-regulate? Will commercial enterprise go too far in their quest to maximise profits? How far is too far? How can organisations ethically make use of face recognition technology to increase efficiencies and drive revenue, whilst respecting and preserving privacy and maintaining the trust of their clientele and society?

Having previously written on the subject of the application of face recognition in airports as applied by law enforcement and border control, this article looks at the increasing exploitation of the technology for commercial advantage. As well as contrasting the different use-cases defined by commercial exploitation versus public safety applications, this article also touches upon the very different agendas of those using the technology and the privacy issues that arise.

1  Advances in Face Recognition Technology

Face recognition is increasingly transforming our daily lives. A study by the US National Institute of Standards and Technology (NIST) in 2010 demonstrated that the technology has improved by two orders of magnitude in accuracy over 10 years and further tests currently being conducted by NIST are expected to demonstrate its continued relentless advance. Those interested in reading of these astonishing improvements are encouraged to refer to “Advances in Face Recognition Technology and its Application in Airports”, first published in Biometrics Technology Today (BTT) in July 2012, which summarises the 2010 NIST results in detail.

2  Public Safety versus Generating Profit

Most people accept that the reality of the world today necessitates certain inconveniences and intrusions. We tolerate and increasingly expect surveillance technology to be deployed wisely in situations where there is demonstrable benefit to public safety, such as at transport hubs, large gatherings, public events or areas of critical national infrastructure. The key factor behind such tolerance is comprehension; we understand the reasoning behind these uses and the benefits to ourselves, namely our safety. Though we don’t necessarily like it, we generally accept it.

However, it has been difficult to avoid the increasing coverage in the media of the use of face recognition by commercial organisations. The single most common term that is bandied about in reference to these deployments tends to be “creepy”. The technology being deployed is very often similar, if not identical to, the technology deployed for public safety applications. So precisely what is it about this use of technology that people are averse to?

In order to understand this, it is useful to consider in each case who people perceive benefit from the system. In the case of public safety, the people perceived to benefit are us; the citizens. In the case of commercial use, people perceive the commercial organisation deploying the technology as the beneficiaries. In this scenario the term “benefit” generally means profit, either by increasing revenues or decreasing costs. Often there is a general distrust within society of large corporations profiting from the exploitation of the populace, and this is especially true in times of prolonged economic difficulty. This is additionally complicated by the fact that our biometric traits are viewed as being something that are intrinsically ours and that are a constituent part of our definition.

3  Examples: Uses to Reduce Cost and Generate Revenue

It hasn’t taken long for business minded technology companies to devise a whole range of new uses of face recognition, all focussed on delivering bottom line business benefit. An important characteristic of face recognition is that it is only useful if you have something to match a photograph (probe) against, whether it is another photograph, or a database of photographs (reference set). It is the management, control of access to and often the creation of these reference sets that generate the most privacy concerns.

Let us briefly discuss some of the manners in which the technology is currently being deployed.

3.1  Efficiently Identifying Customers and Staff

This perhaps is the most traditional use of biometrics within commercial organisations. The ability to positively identify people, whether they are your staff or increasingly your customers, is absolutely necessary for the day-to-day operation of business and indeed society. Biometrics can be applied to ensure identity in a more cost-effective and positive manner, thereby introducing efficiencies into the business. It is an unfortunate reality that staff are responsible for a significant amount of theft. Adopting biometric technology can eliminate password theft and help mitigate the risks of identity sharing, thereby reducing fraudulent and unauthorised transactions and ensuring relevant personnel are physically present at the time of a transaction. Additionally, customers can be identified positively before conducting transactions. Cashless payments provide numerous efficiency opportunities by allowing elimination of cash and credit cards at point of payment altogether.

3.1.1         Privacy Considerations

These examples are usually only possible with the consent and approval of the individuals in question. Customers typically register for a biometric payment system, for example, in order to realise a benefit offered by the enterprise. The enterprise in turn must satisfy the customer that their biometric reference data will be kept and managed securely and only for the stated purpose.

The advent of face recognition provides new manners in which you can identify your customers, for example from CCTV cameras as they enter shops or as they view public advertising displays. It is when these activities are performed without the individual’s knowledge or consent that concerns arise.

3.2  Identifying Who is Entering Your Premises

These solutions are designed to integrate with existing surveillance systems; faces are extracted in real-time from a CCTV video feed and matched against a database of individuals. When the system identifies an individual of interest it can raise an alert that can be responded to rapidly and effectively, or log where and when the individual was seen for the formation of analytical data.

This can be used to provide valuable real-time or analytical intelligence to organisations, such as:retail

  • Notification of the arrival of undesirables, such as banned individuals or known shoplifters.
  • Notification of the arrival of valued or VIP customers.
  • Collation of behaviour data of known customers, such as how frequently they visit, which stores they visit and integration with loyalty programmes.

 

 

 

3.2.1         Privacy Considerations

There are a number of potential issues with regards to privacy that need to be considered here, most notably:

  • How is the reference set obtained? Who is in it?
  • Do you have the permission of the individuals in the reference set?
  • How are the photographs in the reference set stored and secured?
  • Are the members of the reference set aware of how and when their photos will be searched?
  • Are the people crossing the cameras aware that their photos are being searched against pre-defined reference sets?
  • What action is taken if a probe image matches against the reference set? What are the implications of a match or a false match?
  • What is done with the probe images after searching the reference set? Are they discarded or stored?

The number of possible uses of this functionality and resulting business benefits are too large to enumerate here, but very careful consideration must be made with regards to the proportionality of the solution when measured against the requirement. Additionally, the views and considerations of the individuals whose images you are verifying, both the people within the reference set and the people whose faces you are sampling as probe images, should be well understood and considered; approval should be sought for inclusion into a reference set.

3.3  Analysing How People Moving Through Your Premises

Face recognition can also be used to determine how people move through premises, such as a department store. Understanding peak and quiet times is essential to enable sufficient and efficient staffing and resourcing. Raising alerts to manage unforeseen queues is critical for ensuring customer satisfaction.

Face recognition applied to CCTV can timestamp when individuals are detected at known camera locations, thereby providing highly accurate information on people flows such as:

  • How long on average does it take to move between two or more points?
    (such as from the entrance of a store to a checkout or exit)
  • What are the averages flow times across the day and when are the peaks?
  • How does this vary with the time of day?

This can be used to determine how people typically move through the premises, and how long on average they linger in specific areas. You can also analyse this data across different age and gender demographic categories.

3.3.1         Privacy Considerations

Importantly, no person identifying information is recorded. There is no interest in identifying who the individuals moving through the premises are or in taking any specific action on any specific individual. There is no need to search against any pre-defined reference sets.

However, there are some issues you should consider when deploying such systems:

  • Biometric matching of people crossing the cameras still occurs. The probe photos are matched against other anonymous people that have previously crossed the cameras.
  • You should carefully consider how long this data will be retained for matching, (generally hours) and the nature of the premises being monitored.

Generally the privacy considerations of this application are minimal.

3.4  Building Databases of People Visiting Your Premises

As previously mentioned, face recognition is only useful if you have images to match against. Previous examples have dealt with matching the faces of people crossing the camera against known databases of individuals. A potentially far more valuable practice to enterprise is to dynamically build reference databases consisting of the people who cross the camera. Unfortunately, this is also the practice that riles the populace the most and is rife with potential privacy intrusions.

The increase in the use of CCTV cameras has led to an ever increasing volume of archived video footage. The intelligence in this footage typically remains inaccessible unless appropriately analysed and indexed. Such systems can be used to populate databases of “seen” individuals, thereby enabling searching for specific people of interest to determine if, when and where they have been present. This then allows the collation of data such as how frequently individuals visit your premises, how long they stay and when was the last time the individual visited your premises, as well as which of your locations any individual frequents and which is the most common.

If this functionality is combined with the ability to search and cross- reference against databases of known individuals, for example a subscribed customer database, this can then allow you to build very valuable analytical data on specific individuals thereby enabling you to predict future behaviour and market more specific services and products.

3.4.1         Privacy Considerations

Tread very carefully. Some of the most vocal opposition to the application of face recognition technology results from the capture of biometric data of potentially large numbers of people without their knowledge or consent, especially if the people are then identified and profiled against existing databases. In many jurisdictions around the world, the retention of such data may be in contravention of privacy legislation.

3.5  Analysing Who is Viewing What to Target Your Advertising

There have been many examples in recent months of retail and advertising organisations using technology to determine the approximate age and gender of people entering premises or viewing advertising walls. Though not technically face recognition, it is still worth mentioning here as often the distinction between the two uses is blurred. The premise is simple: such solutions can count the number of people watching an advert at any given time, and even estimate their age, dwell time, sex and race. While providing invaluable information for the advertiser, it can also allow them to dynamically change the adverts in real time to more appropriately target the demographic of the current viewer(s). Such solutions are increasingly being deployed in Japan and it is only a matter of time until they are more widely considered in Europe and North America.

3.5.1         Privacy Considerations

The key consideration here is that this form of technology is not actually identifying anybody or extracting personally identifiable information. There does appear to be some opposition to this, though none of it very vocal or serious. It is difficult to see any infringement of privacy and often may be advantageous to the consumer as advertising may be more specifically tailored to their needs.

3.6  Matching People on Your Premises with Social Media Accounts

Both Google and Facebook have acquired face recognition technology companies over the past year. Facebook’s users, for example, publish over 300 million photos onto the site every day, thereby making Facebook the owner of the largest photographic database in the world.

Facebook is already trialling a new service called Facedeals which enables its users to automatically check in at participating retail sites equipped with specially enabled cameras. In order to entice users to participate, the participating retailer can offer special deals to Facebook users when they arrive. The flow of information can be bi-directional. Such automatic check-in data coupled with the users’ manual checkins can be used by Facebook to hone their profile of individuals allowing them to target users with more relevant advertising. The system is entirely voluntarily, and the reference sets searched by retailers only contain photos of users who have opted into the service.

3.6.1         Privacy Considerations

Making data from social media sites available to other commercial organisations is a potential privacy minefield and should only ever be done with users’ consent. Defining these as opt-in services is exactly the right way forward. Likewise the profiling of users of social media sites based upon automatic tagging of images uploaded to those sites should be strictly controlled and only enabled on an opt-in basis. The privacy concerns over such activities have recently been very aptly illustrated by Facebook’s withdrawal of its controversial auto-tagging feature from use in Europe after pressure from privacy campaigners and regulators.

4  Social Media, Cloud Computing and Face Recognition

Dr. Joseph J. Atick of the International Biometrics and Identification Association has written a thought-provoking paper entitled “Face Recognition in the Era of the Cloud and Social Media: Is it Time to Hit the Panic Button?”. The paper raises several interesting points that merit mention here. In it Dr. Atick argues that the convergence of several trends including the:

  • High levels of accuracy now attainable by face recognition algorithms.
  • Ubiquity of social networking with its inherent large photographic databases.
  • Availability of cheap computer processing and the advent of cloud computing.

…coupled with the fact that “face recognition occupies a special place [within the family of biometrics in that] it can be surreptitiously performed from a distance, without subject cooperation and works from ordinary photographs without the need for special enrolment…” is “ … creating an environment … that threatens privacy on a very large scale…”.

One of the main premises of the paper is that this issue “… will require the active cooperation of social media providers and the IT industry to ensure the continued protection of our reasonable expectations of privacy, without crippling use of this powerful technology”.

5  Can All This be Done Ethically? (What About Privacy?)

Can organisations ethically make use of face recognition technology to increase efficiencies and drive revenue, whilst respecting and preserving privacy and maintaining the trust of their clientele and society?

The premise of “privacy-by-design” should be used to ensure that privacy is considered from the outset of any deployment of face recognition technology. In fact, the European Union’s 22-month Privacy Impact Assessment Framework (PIAF) project advises that “Privacy impact assessments should be mandatory and must engage stakeholders in the process” for all biometric projects.

Reputable organisations such as the Biometrics Institute have gone so far as to publish invaluable privacy charters to act as a “…good executive guide operating over a number of jurisdictions…” which should be reviewed and seriously considered before any deployment of biometric technology.

Some of these fundamental principles are outlined below within context of the subject matter of this article and specifically within the context of commercial use of the technology. These will not necessarily apply when discussing matters of public safety, law enforcement and national security.

5.1  Proportionality

A fundamental principle of privacy concerns the limitation of the collection of data to that which is necessary. Organisations should not collect more personal information than they reasonably need to carry out the stated purpose. Biometric data by its very nature is sensitive and absolute assurance must be provided that it will be managed, secured and used appropriately. However, a key consideration in the use of this technology should be proportionality; is the collection of such sensitive data justified for the benefit realised?

5.2  Educate and Inform

People on the whole generally resent not being informed, especially in matters that involve them. History is littered with IT projects that have failed because key stakeholders were not involved from the outset, were not sufficiently informed and whose buy-in to the process was not obtained. Customers are one of the most important stakeholders and these issues are even more critical when dealing with their personal and biometric data.

There is a very interesting video on YouTube that illustrates this point very nicely. It is filmed by a man with a camera walking around filming random strangers without explanation. The reaction is predictably always negative and sometimes hostile. The message the video is trying to make is obvious: most people do not approve of being videoed, so why do we so readily accept surveillance cameras? The message that comes across is actually clearer: People object when they do not understand intent, purpose or benefit to themselves. The cameraman offered no messages of explanation of his intent, even when challenged. Objection was guaranteed.

5.3  Be Truthful and Accurate when Describing the Business Purpose and Benefit

As part of the process of informing, organisations should also be direct and open in disclosing not only the existence of the systems, but the scope, intent and purpose of the solutions. Why are you utilising an individual’s biometric data? What benefit does it serve? What is the scope of the use of this data?

Importantly stay well clear of “scope creep”. All too often it is tempting to start using data once you have it for other than the stated intended purpose for which it was collected. Such endeavours will inevitably lead to loss of trust.

5.4  Provide Benefit to the Customer

Simply understanding the scope, purpose and intent of a system generally will not be sufficient to garner acceptance of the system. While people are generally astute enough to realise that businesses are in the business of making money, they’ll want to know what is in it for them. What is their benefit?

An example with which most of us will be familiar are grocery store loyalty or “club” cards. Whilst we all understand the objective of the grocery store is to profile and analyse our spending in order to better market to us, a majority of us still subscribe in order to receive the enticements and benefits on offer.

Within the context of face recognition, Facebook’s Facedeals programme demonstrates this principle nicely. Users understand the benefit to Facebook and the retailer, yet they still may choose to opt in to the programme because there is a clear and discernible benefit for them to do so as well, namely targeted discounts and offers at retail outlets.

This is also affirmed by a survey in 2012 by IATA which finds that “… most travellers are receptive to the idea of using biometrics within the border control process.” Why? Because there is clear and discernible benefit to them in the form of a more efficient passenger process and increased levels of security.

5.5  Seek Consent and Operate on an Opt-in Principle Where Appropriate

Biometric enrolment into such systems should not be mandatory. Individuals should be allowed the ability to opt-in, with an opt-out status being the default. Clearly this is not always feasible when considering people in public places the crossing cameras. However, if they are being identified against reference sets, the individuals in the reference sets should be there only with consent. Automatic enrolment into reference sets or biometric databases should involve the consent and approval of those enrolled.

Importantly, people should not be penalised should they choose not to opt-in; they should still be allowed a mechanism of transacting and conducting their business.

The recent decision by the UK Department of Education to prohibit schools from taking pupils’ fingerprints or other biometric data without gaining parents’ permission is a prime example of a potential backlash when such systems are made mandatory without providing any alternative mechanism of transacting. In many cases in UK schools, students were left with no mechanism of buying their school lunch unless they enrolled into a biometric system.

6  Summary

The accuracy of face recognition has increased dramatically. Retailers and other commercial organisations are investigating ways to exploit this technology to increase revenues, improve margins and enhance efficiency. Social media companies own the largest photographic databases in existence and are under pressure from shareholders to find ways to monetise these assets. As these explorations gather pace, so does the discontent of privacy advocates.

This article has outlined a number of ways face recognition can be used by enterprise and highlights potential privacy issues. Is it possible to ethically use face recognition technology and respect privacy? This will only be possible if enterprise maintains the trust and respect of its customers. Open and honest discourse is the best manner in which to achieve this. This should be accompanied by delivering real benefit to all parties involved in a manner that also empowers the customer; nobody should be forced to enrol into biometric systems or be disenfranchised from refusing to do so.

How far is too far? History has shown that there is no absolute answer to such questions. The exact location of the line to be crossed is always a factor of and changes with the times we live in. History has also shown, especially as it pertains to technology, that it is next to impossible to put the genie back into the bottle once released. It is now the collective responsibility of all to ensure the proper and ethical use of this technology in a manner that delivers the maximum benefit. This will require the active cooperation of social media, enterprise, the IT industry and civil liberty groups to ensure the continued protection of our reasonable expectations of privacy without crippling the use of this powerful technology. In the end, the people have the loudest voice. If enterprise crosses the line, customers will pass judgement with their wallets. 

7  About the Author

Carl is the founder of Allevate Limited (http://allevate.com), an independent consultancy specialising in market engagement for biometric and identification solutions. With over 20 years’ experience working in the hi-technology and software industry globally, he has significant experience with identification and public safety technologies including databases, PKI and smartcards, and has spent the past 10 years enabling the deployment of biometric technologies to infrastructure projects. Carl started working with biometrics whilst employed by NEC in the UK and has subsequently supported NEC’s global and public safety business internationally.

Residing in the UK, Carl was born and raised in Canada and holds a Bachelor of Science Degree on Computer Science and Mathematics from the University of Toronto.

You can download a PDF copy of this article by clicking this link.

 

3,976 words

 ————————————————————————————————————————

[i] http://biometrics.nist.gov/cs_links/face/mbe/MBE_2D_face_report_NISTIR_7709.pdf
Multiple Biometric Evaluation (2010) Report on Evaluation of 2D Still Image Face Recognition
Patrick J. Grother, George W. Quinn and P. Jonathon Phillips

[ii]http://allevate.com/blog/index.php/2012/07/17/advances-in-face-recognition-technology-and-its-application-in-airports/
Advances in Face Recognition Technology and its Application in Airports
Carl Gohringer,  Allevate Limited,
July 2012

[iii]http://www.ibia.org/download/datasets/929/Atick%2012-7-2011.pdf
Face Recognition in the Era of the Cloud and Social Media: Is it Time to Hit the Panic Button?
Dr. Joseph Atick
International Biometrics and Identification Association

[iv] http://www.piafproject.eu/

[v] http://www.biometricsinstitute.org/pages/privacy-charter.html
Privacy Charter
Biometrics Institute

[vi] http://www.iata.org/publications/Documents/2012-iata-global-passenger-survey-highlights.pdf
2012 IATA GLOBAL PASSENGER SURVEY HIGHLIGHTS
The International Air Transport Association (IATA)


Article: Advances in Face Recognition Technology and its Application in Airports 1

You can download a PDF copy of this article by clicking this link.

The accuracy of face recognition has increased dramatically. The top performing algorithm in independent evaluations by the US National Institute of Standards and Technology (NIST) is now capable of providing reliable results in real-world environments; the technology is being deployed in airports today to enable everything from automated immigration processes, improved surveillance, security and seamless passenger travel, to the gathering of valuable statistical information pertaining to passenger movements.

1.0 The Business Environment

Airports are complex environments involving multiple stakeholders with conflicting requirements:

  • Government and border control.
  • Police.
  • Airport operators.
  • Airlines.
  • Retailers.

All parties must comply with all Government regulations and utilise the latest documents and passports from multiple issuing states while adhering to all security requirements.

1.1 More Passengers, Same Resources

Passenger numbers are relentlessly increasing; border crossings into the European Union by air alone are expected to increase to 720 million by 2030. The need to mitigate risk is constantly weighed against the requirement to ensure passenger mobility, whilst accurately and unambiguously identifying all those who move through this complex environment.

Biometrics is playing an ever-increasing role in response to these multi-faceted requirements.

2 Advances in Face Recognition Technology

Enter Face Recognition biometrics. This technology is set to transform CCTV surveillance. It is here. Ready to deploy. Now. A recent study by the US National Institute of Standards and Technology (NIST) demonstrated that the accuracy achieved by the top vendor can provide clear and measurable benefits to a range of applications, including surveillance.

2.1 Order of Magnitude Improvements between Subsequent Tests

Most remarkable is the rate of improvement in the accuracy of face recognition algorithms. NIST testing has demonstrated an order of magnitude of improvement in the False Reject Rates (FRR) every four years. Whilst maintaining a False Accept Rate (FAR) of 0.001, the FRR over 3 tests spanning 8 years were:

  • 2002: FRR of 0.2
  • 2006: FRR of 0.01
  • 2010: FRR of 0.003

Put simply, in the latest tests, if the best performing algorithm was set so that it would not falsely match two images of different people more than once in every 1,000 attempts, it would then fail to match two images of the same person only three times in every 1,000 attempts. In contrast, the 2002 tests mismatched 20 times for every 100 attempts.

FRR Results from NIST MBE2010

This arguably outperforms the accuracy of human beings.

2.2 Laboratory Testing versus Real World Environments

Although the above results are excellent, the controlled conditions of a laboratory environment are not representative of real-world conditions. They are a good indicator of results that may be attained when comparing photos of a similar quality taken under similar conditions (i.e. identifying 1 passport photo against a database of passport photos). However, photographs taken in an automatic border control e-gate or from a CCTV camera are not taken under the same control. These are commonly termed non-compliant captures.

Traditionally, face recognition software suffers degradation in accuracy when dealing with challenges such as variable lighting conditions or non-frontal images of the subject. Vendors that can better deal with these challenges deliver systems that perform in a consistently more reliable fashion in the field.

The latest NIST test indicates that the ability of the software to deal with the challenges of non-compliant photos has drastically increased. Face recognition software can now be reliably deployed in airport environments to deliver real and tangible business benefits.

2.3 Increased Tolerance to Angle / Pose

One way to predict how well a face recognition algorithm will perform in a real-world environment when dealing with non-compliant captures is to measure how well it performs in the laboratory with non-frontal photographs (where the subject‘s image is captured at an angle).

These lab results are an indicator as to which solutions will perform better when applying face recognition to CCTV cameras.

The recent NIST tests showed that the most accurate algorithm is highly tolerant to changes in pose. This indicates that detection rates from CCTV cameras should provide tangible benefits whilst minimising the level of false alarms.

Photos at a 25 Degree Angle

2.4 Increased Tolerance to Time Between Photographs

Additionally, it is often the case that the reference photographs we are comparing the live captures to are not recent. For example, most passports are valid for 10 years, so it is essential that we can still maintain a high level of accuracy when verifying photographs against older reference sets.

The NIST MBE 2010 study demonstrated that the highest performing algorithm was able to maintain accuracy rates that deliver quantifiable benefits in these circumstances.

Photos Taken 8 Years Apart

2.5 Lower Resolution Photos

It is also common for non-compliant face captures from CCTV cameras to involve photographs in which the subject’s face constitutes a small percentage of the overall frame of the picture or where the face resolution is not particularly high. This may be due to the use of a lower resolution camera or due to distance between the subject and the camera.

In December of 2011, NIST published another report entitled The Performance of Face Recognition Algorithms on Compressed Images. Although not the primary driver of this study, the results clearly show that the same top performing algorithm was able to generate the same high levels of accuracy with inter-eye distances all the way down to 24 pixels between the eyes, thereby providing another indicator of expected accuracy in real-world environments.

2.6 Real-World Results

There are numerous factors in a live deployment that need to be considered such as lighting, camera postion, distance of the subjects from the camera and the angles at which the sample photographs are taken.

Recently, a national government conducted an evaluation of an e-gate solution at an airport. As part of this evaluation, e-passport holders were invited to use the gates. A sufficient number of passengers were subsequently processed through the gates to provide proper statistical significance. Algorithms from three separate face recognition vendors were tested in the gate.

In this real-world scenario, passport photos of passengers were verified against a lesser quality livescan photo taken within the e-gate itself. The results were presented at the Biometrics 2010 Conference in London: the top performing vendor in the NIST test was able to achieve a real-world FRR rate of 1.1%. This is arguably a better result than can be obtained by a live border guard manually comparing passport photos against the passport holders.

FRR Attained in a Live immigration e-Gate

3 Current Applications of Face Recognition in Airports

Face recognition has evolved significantly over the past decade and has now attained a level of accuracy that provides real and quantifiable business benefit to all stakeholders in an airport environment. Solutions incorporating face recognition are already being deployed today.

3.1 Automated Border Control Gates at Immigration

Many nations world-wide have deployed e-Passports which are being carried by an ever-increasing percentage of the world’s population. This enables governments to deploy Automated Border Control (ABC) gates. In EU nations for example, these gates:

  • are for EU passport holders only.air travel
  • do not require pre-enrolment.
  • perform a 1:1 face verification of a live scan against the JPG on the passport chip.

In the UK these gates are being widely deployed at entry ports and seemingly form the backbone of the government’s strategy for automatically clearing EU passengers.

In Asia the three largest ABC deployments in the world (Singapore, Macau and Hong Kong) each process hundreds of thousands of passengers daily, maximising the efficiency of live border guards.

3.1.1 How it Works

The process involved in an ABC gate is fairly simple:

  1. The passenger approaches the gate and has their passport read by the e-gate.
  2. The validity of the data page on the passport is verified using a variety of tests.
  3. The information in the machine readable zone (MRZ) is verified against the data read off the chip.
  4. The passport information is sent to the appropriate government systems for the appropriate checks.
  5. If there are any problems thus far, the passenger is re-directed to a manned border lane, otherwise …
  6. A live photo is captured of the passenger (with appropriate liveness checks).
  7. Face recognition is used to verify the live capture with the photograph read off the passport’s chip.
  8. If the photo does not match, the passenger is assisted by a live border guard, otherwise…
  9. The passenger is allowed to proceed.

In this use of face recognition:

  • FAR represents the percentage of passengers holding a passport that does not belong to them that are wrongly admitted.
  • FRR represents the percentage of legitimate passengers who are wrongly re-directed to a live border guard due to the photographs not matching.

There have been no published studies of the FAR and FRR achieved by a live border guard, but it is generally accepted that face recognition operates at a higher level of accuracy, especially when a border guard has been on operational duty for more than 2 hours or has to deal with visual verification of multiple races of passengers. Most e-gate deployments in Europe today operate with an FRR of approximately 6% set against a corresponding FAR of 0.1%.

Recently, an officer responsible for a large deployment of e-gates in an international airport indicated that in his view, most imposters attempting fraudulent entry into the country prefer to try their luck with manned border guards rather than use automated gates.

3.1.2 The Business Benefit

You don’t have to look far today to read of the burgeoning deficits of most western nations. Austerity is the order of the day. Even in light of the expected year-on-year growth in passenger numbers, budgets are being cut. More and more often, improved efficiencies introduced by the sensible deployment of technology are being relied on to address these budget shortfalls.

Border guards are highly skilled and experienced staff deployed at the front-line of our nations defences. 99% of travellers entering a country are benign. Routine checking of travel documents and verification of valid ownership are tasks that can now be better performed by technology, thereby enabling the automated egress of legitimate travellers and allowing the border guards to focus on and find the 1% of the travellers they really want to speak with. In effect, removing the haystack to reveal the needle.

It is also relevant to note that the higher the accuracy of the face recognition solution deployed, the lower the FRR realised, thereby resulting in fewer passengers redirected to a live border guard and a lower cost of total ownership.

3.1.3 An Example

Another nation that has recently trialled the deployment of 4 ABC lanes determined the following:

  • Without the ABC lanes, 8 manual lanes required 8 border guards.
  • With the ABC lanes, the same 8 border guards were able to monitor 12 lanes.
  • Without the 4 ABC lanes, 8 border guards oversaw the entry of 950 passengers per hour.
  • With the 4 ABC lanes, 8 border guards oversaw the entry of 2,400 passengers per hour.

Even with the deployment of a limited number of ABC lanes a real and tangible benefit was realised.

3.2 Trusted Traveller Systems

Most ABC solutions deployed today take one of two forms:

  • Non-Registered, for holders of e-Passports from authorised countries (as discussed above).
  • Registered, for holders of passports from countries not authorised to use the Non-Registered lanes (or holders of older passports without a chip).

Examples of the latter include the US Global Entry, Dutch Privium (collectively FLUX) and the UK IRIS systems.

As non-registered systems become more commonplace and the number of e-passport holders continues to rise, the business case for governments to provide separate free-to-use Trusted Traveller systems becomes vague. Ideally, given the limited space available in airports, the best scenario involves these passengers using the same physical e-gates as users of the non-registered systems.

Existing e-gates can be modified to accommodate holders of e-Passports from other nations. An additional step in the process flow allows the e-gate to cross-reference against a database of pre-enrolled and vetted Trusted Travellers. An additional face verification can be performed against the stored face details of the enrolled passenger.

3.3 Departure and Boarding Gates

The previous example depicts the use of biometrics to facilitate passenger processing at immigration and to introduce efficiencies to the tasks of border control officials. Airport operators and airlines are also increasingly turning to biometrics to facilitate the flow of outbound passengers through airport terminals.

Simplifying Passenger Travel (SPT) was an initiative led by airlines, airports, governments and technology providers which proposed the “Ideal Process Flow”. The goal was to combine e-passports, biometrics and network infrastructure to enable the automatic identification and processing of passengers to move them through the airport seamlessly while freeing up staff to concentrate on security threats and customer service.

While the full ambition of assigning a single biometric identifier to a passenger’s entire airport journey, from booking, to check-in, bag drop through to security and eventually boarding is yet to be realised, key elements are already being implemented by airport operators.

3.3.1 The Problem

Many airport terminals have a single common departure lounge for both domestic and international passengers. Here exists the potential for a departing domestic traveller to swap boarding cards with an arriving international traveller, thereby enabling the arriving traveller to transit to a domestic airport and bypass immigration processes.

3.3.2 The Solution

This problem can be remedied by introducing automatic gates with face recognition at the entry to the common departure area and at the gate prior to airplane boarding. The automated gate at plane boarding captures the passenger’s face and verifies it with the face captured and associated with the boarding card when the passenger entered the departure area, thereby detecting if a boarding card has been swapped.

3.4 Surveillance: Real Time Watchlist Alerts

Matching faces captured from CCTV against photographic databases has long been the holy grail of face recognition. These systems are now being deployed today.

Although the results obtained in the NIST evaluations do not reflect the results that can be obtained in a live surveillance environment, it stands to reason that solutions that incorporate the best performing algorithms will also yield the highest accuracy results when matching CCTV images against a watchlist.

3.4.1 What it Delivers

These solutions are designed to integrate with existing surveillance processes; faces are extracted in real-time from the CCTV video feed and matched against a watchlist of individuals. When the system identifies an individual of interest, it raises an alert that can be responded to rapidly and effectively.

In this application of face recognition:

  • FAR represents the percentage of people captured by a CCTV camera that are falsely matched against the watchlist (in essence the number of false alarms raised by the system).
  • FRR represents the percentage of people captured by a CCTV camera who are in the watchlist but for which no alarm is raised.

The alerting mechanism is a binary process. If the system raises too many false alarms, it will quickly be ignored by those tasked with responding to these alerts. The objective of these systems is to minimise the false alerts to a manageable level, while detecting the highest possible percentage of people moving past the cameras who are in the watchlist (true ID rate).

3.4.2 Challenges

It is essential that expectations are set appropriately. Scenarios where thousands of cameras are scanning large crowds of people in day and night environments and from a distance to identify individuals of interest are still largely unrealistic. The best results are obtained:

  • Using newer high definition cameras (3-5 megapixels).
  • Indoors with uniform lighting or outside during daylight in the absence of specific glare.
  • Where people are generally facing the same direction and moving towards the camera.
  • In a suitable pinch-point, such as in a corridor, lane or access gate / turnstile (not large crowds of people).
  • Where cameras are positioned in such a manner as to minimise the angle to the face (ideally < 20 degrees).

Additionally, as the system is comparing poorer quality photos captured from CCTV, it is imperative that the highest quality reference photos are inserted into the watchlist. Systems comparing poor photos against poor photos operate at significantly reduced accuracy levels.

Even with the above considerations in mind, there exist substantial opportunities and environments in which these solutions may be deployed to deliver significant results.

3.4.3 Technical Considerations

These solutions are typically deployed in environments where large numbers of people may be crossing the cameras. As such, depending on the size of the watchlist, a very large number of face verifications need to take place. Such solutions potentially require intensive use of server infrastructure.

Typically, the main considerations that determine the server infrastructure required are:

  • The size of the watchlist.
    (Typically, these would only contain key or significant individuals.)
  • The number of people moving across the camera(s).
    (This represents the number of transactions or searches against the watchlist.)
  • The response time required in which to raise an alert.
  • The number of frames per second which are being captured by the cameras.
    (The higher the frame rate, the more times you capture the same person walking past the camera.)

Real-time searching of an entire criminal database is not typically feasible; consideration should be taken when determining who should be inserted into the watchlist to minimise its size. Typical watchlist sizes are in the hundreds or thousands.

The two major areas of processing inherent in such a system include:

  • Creating biometric templates of all the faces moving across the CCTV camera.
  • Matching these biometric templates against the watchlist.

Of these, template creation generally requires the most CPU power and time.

Therefore very careful consideration must be given to the number of frames per second (fps) the cameras are running at. Many systems typically run at 5-10 fps. While the processing power is significantly reduced, so is the overall accuracy of the system. The lower the fps, the more likely it is that the system will throw away frames containing a high quality image of the individuals’ faces.

To obtain optimal accuracy, cameras should be running at up to 20fps. However, this will result in more images of the same person being captured, resulting in a higher level of template creations and searches.
Solutions must be designed with scalability in mind, allowing the most efficient use of server power available.

3.4.4 An Example

An example of an existing live deployment in an airport environment consists of:

  • Up to 10 five megapixel cameras running at 25 fps.
  • A peak transaction rate of 1,000 people per minute moving across the cameras.
  • A watchlist of up to 1,000 people.
  • An alert response time of 5 seconds.

Each person is captured tens of times, resulting in tens of thousands of template creations per minute and tens of millions of biometric verifications per minute.

In this environment, assuming suitable environmental conditions and positioning of the cameras, this system identifies people in the watchlist up to 90% of the time (true id rate) with only one false alarm per day. If operators are willing to accept more false alarms, the true id rate can be increased by configuring system tuning parameters and lowering matching thresholds.

Such systems are already running today.

3.5 Surveillance: Forensic Video Analysis

The increase in the use of CCTV cameras has led to an ever increasing volume of archived video footage. The intelligence in this footage typically remains inaccessible unless appropriately analysed and indexed. Reducing investigation hours when limited resources are available is essential. Such systems can be used to populate databases of “seen” individuals, thereby enabling authorities to search for specific people of interest to determine if, when and where they have been present.

3.5.1 How it Works

  • Faces of individuals are captured from CCTV and archived in a database.
  • Authorities can search the archive using a photo to determine a camera ID and timestamp.
  • Playback of the relevant recording can be enabled by storing pointers into the video archive.

3.5.2 Usage Example: Passengers without Documentation

One usage already deployed today is to quickly and accurately determine the point of origin of arriving passengers without documentation, such as asylum seekers.

If a passenger presents themselves to immigration without documentation and does not provide accurate or complete information about themself, authorities can capture a photograph of the person and search the database of archived faces. If cameras are placed in aerobridges to record disembarking passengers, it is then a simple process to identify on which flight the passenger arrived.

3.6 Queue Management and Flow Analysis

It is becoming increasingly important for airlines and airport operators to monitor queue lengths and passenger flows within the airport. Understanding peak and quiet times is essential to enable sufficient and efficient staffing and resourcing. Raising alerts to manage unforeseen queues is critical for ensuring passenger satisfaction as well as for ensuring that all SLAs with other stakeholders, such as airlines or government agencies, are adhered to.

A common solution thus far has involved the tracking of bluetooth enabled devices such as PDAs and smartphones which are carried by passengers. However, relatively low percentages (approximately 15%) of passengers carry such a device, let alone have the bluetooth on the device activated.

A solution that provides a much more comprehensive data set and accurate information is needed.

3.6.1 The Application of Face Recognition

Solutions using CCTV with face recognition can timestamp when individuals are detected at known camera locations, thereby providing highly accurate information on passenger flows such as:

  • How long does it take to move between two or more points? (such as check-in to security)
  • What are the averages and when are the peaks?
  • How does this vary with time of day?

…as well as providing invaluable insight on how passengers move through the airport:

  • What percentage of passengers move from security to duty free?
  • How many of these are male / female?
  • How long does the average passenger spend shopping in duty free?
  • How is this impacted by queue lengths?

Importantly, no specific passenger identifying information is recorded.

3.6.2 How it Works

As passengers enter an area of interest they are acquired by a camera and anonymously enrolled into the system:

  • CCTV cameras enabled with biometric technology are installed at appropriate areas of interest.
  • Passengers are automatically searched against the database of enrolled individuals.
  • The passenger’s anonymous record is updated with a camera number and timestamp.
  • The database is automatically purged as required at regular pre-defined intervals.
  • The system can raise the appropriate alerts as required (i.e. queues too long).

4 Privacy Considerations

Any article on face recognition would be seriously remiss without at least mentioning privacy. There are a multitude of sources available for detailed discussions on privacy versus benefit of this technology, including the views of this article’s author; readers should familiarise themselves with this issue before considering any deployment of face recognition.

5 What’s Next?

As the use of face recognition continues to be substantiated, more ingenuitive applications will be deployed. Cloud-based services will enable the transfer of expensive computing power out of the airport into shared server facilities. Face recognition will assign a passenger a single unique and transient identity during their movement through the airport, thereby allowing them to be processed by multiple applications seamlessly and effortlessly. Passenger movement through an airport environment can be tracked up to the point of their departure. Personalised way-finding solutions can guide individual passengers to their specific gate, thereby reducing flight delays and passengers who are delaying flights can be quickly and easily located.

6 Summary

The accuracy of face recognition has increased dramatically over the past years. The top performing algorithm in independent evaluations by the US National Institute of Standards and Technology is now capable of providing reliable results in real-world environments and the technology is being deployed today in airports to enable everything from automated immigration processes, improved surveillance and security, seamless passenger travel and the gathering of valuable statistical information pertaining to passenger movements. The number of potential applications of this technology will continue to deliver benefits in creative ways we have yet to imagine.

The business benefit is real and quantifiable.

7 About the Author

Carl is the founder of Allevate Limited (http://allevate.com), a consultancy focused on providing strategic expertise for identity projects that incorporate biometrics, automation and analytic technologies. With over 20 years’ experience working in the hi-technology and software industry, he has spent the past 10 years enabling the deployment of biometric technologies to national infrastructure projects. Carl started working with biometrics whilst employed by NEC in the UK. Allevate continues to work closely with NEC on identification projects in Europe for government, border control and law enforcement.

You can download a PDF copy of this article by clicking this link.

This is the author’s original version of a work that was accepted for publication in Biometrics Technology Today (BTT). Changes resulting from BTT’s publishing process are not reflected in this original version, and as such this article may differ from the version subsequently published in Biometrics Technology Today, VOL: 2012, ISSUE: 7, Date: July, 2012,  DOI: http://dx.doi.org/10.1016/S0969-4765(12)70148-0