VIEW CONES How Consensual is Our City?
Diploma Unit 3
Lingge Yang TS 5 2018-19 Christina Varvia + Merve Anil
In August 2016, Metropolitan Police Firstly deployed a Facial Recognition trial at Notting Hill Carnival with a single monitoring van. Following the trial, a heated debate about its intrusion of privacy and discriminatory impact on certain group of people has raised. Nowadays, CCTV surveillance has formed a contemporary Panopticon in civil society, and recently developed Facial Recognition technology has facilitated this further with more efficient data processing capacity, which means individual’s identity can be recognised and location can be tracked. As a method of managing crimes and ensuring security, surveillance is portrayed as necessary to maintain a state’s political unity and sovereignty. For an ordinary individual, it is a social contract that one has to subscribe to gain civil rights. The way that surveillance is practicing in urban context is that people have no choice to subject to it or not as CCTVs are always located at points where social/ public services and city’s infrastructure are provided. To address this issue, the project propose to visualise the surveillance-scape in the city by mapping out CCTV view cones and areas visible through CCTV system. The proposal is a specialised version of GDPR, which empowers citizens with an agency that asks their consents when they step into watching areas.
Content Chapter 1 Introduction 1.1 How Consensual is Our City? 1.2 Ask Consent and Privacy by Design - Spatialising GDPR 1.3 Investigation Methodology - Seeing through City's Eyes
Chapter 2 What is a Consensual Space 2.1 CCTV Disperse in City - Social Contract 2.2 Private and Public Space - Redefining through Technology 2.3 Accessibility / Opacity - CCTV Surveillance Operation System
Chapter 3 Notting Hill Carnival Facial Recognition Trial 3.1 Police’s Deployment of Facial Recognition Trial 3.2 INVESTIGATION: Interview and Scene Reconstruction 3.3 Notting Hill Carnival Operation Breakdown 3.4 Main Factors - Algorithm - Camera Feeds - Watch List - False Positive 3.5 Intrusion of Privacy - Lack of Legal Framework
Chapter 4 CCTV 4.1 Evolution of CCTV 4.2 Public CCTV - Conduct Code 4.3 Private CCTV - Conduct Code 4.4 Types of CCTV Camera 4.5 TEST 01: Specs Affecting Capturing Space - Focal Length & Resolution - Visualising View Cone 4.6 Data Storage and Retention 4.7 TEST 02: Specs Affecting Capturing Activities - Frame Rate & Resolution - Mapping Activities 4.8 Data Retention Protocol 4.9 Control Room Design 4.10 View from Control Room - Fragmented City 4.11 TEST 03: View from Control Room - CCTV View Range in Euston Station - Interior Room CCTV Analogue - Photogrammetry Test
Chapter 5 Facial Recognition 5.1 Facial Recognition Overview 5.2 Deep Neutral Network 5.3 Facial Recognition Framework 5.4 Applications of Facial Recognition 5.5 Principles and Coding Structure 5.6 TEST 04: Fabricating Identities - Database Structure - Real Identities - Exchange Identities - Fake Identities 5.7 TEST 05: Architectural Elements Affecting Facial Recognition - Distance and Perspective - Lighting Level - Blockage / Barriers - Material - Translucency - Reflection
Chapter 6 Contextualisation 6.1 TEST 06: Sculpting Space - View Cones Mapping - Activities Capturing - CCTV's View
Chapter 7 Threshold 7.1 New Cartography of the City 7.2 Shaping Behaviours with Shades 7.3 Design from Behaviours 7.4 Conclusion: Threshold
Chapter 8 Bibliography
Chapter 1 Introduction 1.1 How Consensual is Our City? 1.2 Ask Consent and Privacy by Design - Spatialising GDPR 1.3 Investigation Methodology - Seeing through City's Eyes
HOW CONSENSUAL IS OUR CITY? As a method of managing crimes and ensuring security, surveillance is portrayed as necessary to maintain a state’s political unity and sovereignty. For an ordinary individual, it is a social contract that one has to subscribe to gain civil rights. The way that surveillance is practicing in urban context is that people have no choice to subject to it or not as CCTVs are always located at points where social/public services and city’s infrastructure are provided. If there is any form of accountability and process of approval of surveillance before it can be implemented, asking "How Consensual is Our City" is acturally questioning whether one can live in a city without giving consent to CCTVs.
CCTV Density in Greater London Number of cameras per square mile More than 20 10-20 Fewer than 10 No data
It is estimated there are
Between 2007 - 2010
The average Londoner is
500,000
£300
CAUGHT ON CAMERA
CCTV cameras in London
pent on CCTV installation
times a day
MILLION
300
.9.
Spatialising GDPR
ASK CONSENT GDPR demonstrated an ethical understanding of data surveillance and its potential of abuses. It regulates that the request for consent must be given in an intelligible and easily accessible form, with the purpose for data processing attached to that consent. Consent must be clear and distinguishable from other matters and provided in an intelligible and easily accessible form, using clear and plain language. It must be easy to withdraw consent as it is to give it. In the case of CCTV surveillance, practice code regulates that a clear sign should be given where CCTV is operating and capturing public spaces. However, it is not intelligible for people to know where the watching area starts and ends, and in many cases there it is not acknowledged that who is operating the CCTV and how much information is captured by the camera. Visualising the view cones and other information behind cameras is providing people an agency to make decision on their own.
. 10 .
https://eugdpr.org/ (Eugdpr.org, 2018)
Spatialising GDPR
PRIVACY BY DESIGN Privacy by design as a concept has existed for years. But it is only just becoming part of a legal requirement with the GDPR. As its core, privacy by design calls for the inclusion of data protection from the onset of the designing of systems, rather than an addition, so as CCTV surveillance being practiced spatially. How do we design space by considering consensual as an element in the first place? The visualisation of view cones is using virtual boundaries to sculpt space with vision thresholds. This methodology could inform how to design a consensual space. Consensual space does not equal to private space, on the contrary, a consensual space is a way to redefine public space, where individuals are empowered to choose whether give out their consent or not by clarify the boundaries. In the case when consent is given, the CCTV and facial recognition can be effectively deployed to collect data of human behaviours within the space and the data then can be analysed for further design purpose.
https://solid.inrupt.com/how-it-works (Solid.inrupt.com, 2019)
. 11 .
Two closed-circuit television system operators monitor Islington’s control room, where they can watch images from the borough’s extensive camera network. (Draper, 2018)
. 12 .
Seeing through City's Eyes
INVESTIGATION METHODOLOGY
The principle of the investigation methodology for this technical study is to read space and human activities through CCTV and Facial Recognition systems. The study starts from Notting Hill Carnival Facial Recognition trial as a case study. By reconstructing and unpacking the operation scene, the study diagnoses technological and operational factors affecting the final results. Each factors are then broken down with technical parameters and legal frame work to analyse how this operation was intrusive and discriminatory. The study then looks into technical details of CCTVs and Facial Recognition as hardware and software of the surveillance system. Along research, tests are taken to demonstrate how does each parameter affect capturing and interpreting spatial information. The mapping language is developed long through research and tests by cataloguing and classifying different parameters. The study is wrapped with contextualising all the research and testing results in a real space as a visual demo of the proposed mapping system. The study proposes a new cartography of city, which could potentially affect the way people interfaces with city. It comes up with a propositional speculation that use surveillance and machine learning as a way to collect data and generate architectural design.
. 13 .
. 14 .
Chapter 2 What is a Consensual Space 2.1 CCTV Disperse in City - Social Contract 2.2 Private and Publics Space - Redefining through Technology 2.3 Accessibility / Opacity - CCTV Surveillance Operation System
Where is the Most Surveilled Area in the City?
CCTV DISPERSE IN CITY - SOCIAL CONTRACT
Fixed Public CCTV locations mapped are based on a Freedom of Information request to Camden Borough Council which was answered on 1st February 2011. The pattern shows that CCTV are more likely to be located on high street and around public transport stations. As an ordinary people living in the city, being watched by CCTV always occurs when one goes to those public spaces and participate in the civil society. You are being watched when you are using this city’s infrastructure. You are being watched when you function as a part of the society. You are being watched to exist. It is not merely that some kinds of surveillance may seem invasive or intrusion, but rather that social relations and social power are organised in part through surveillance strategies. CCTV is not just a measure for crime management, but a social contract that everyone has to subscribe to in exchange of the privilege to be part of the society and able to use public services.
. 17 .
. 18 .
Fixed CCTV Locations are mapped from a Freedom Information Request https://www.whatdotheyknow.com/request/cctv_locations_2
CCTV Cluster around High Street
CCTV Cluster around Train Station
CCTV Cluster around Train Station
CCTV Cluster around High Street
. 19 .
Redefine Private and Public Space through Technology
PRIVATE AND PUBLIC SPACES
The technology development does change our perception of space and challenged the traditional distinct of public and private spaces. When a physical space is purely observed by attendance, the perception is limited by observer’s spatial and temporal existence. The use of CCTV enables certain public spaces and human activities in the space being watched constantly and remotely by the institution. In the situation when everyone has a camera, the perception of the space has been changed as photos or videos might be taken in spaces not accessible to the public, so we can have an image of a space even we are absent from there. Moreover, live streaming and webcast makes it possible to not only portray the space but also watch the people’s behaviour inside those spaces. What facial recognition brings it further is, not just subjects being recorded but being identified lively, so individual can be located and tracked through this setting by the agent who’s running the system. So we can speculate a future scenario, just like camera and live streaming, what if the power of recognising is decentralised, so everyone has the ability to recognise each other lively.
. 20 .
. 21 .
. 22 .
. 23 .
CCTV Surveillance Operation System
ACCESSIBILITY / OPACITY
The urban CCTV surveillance system is investigated through five places covering different spectrum of private and public. Boundaries of public accessibility and CCTV camera locations in these buildings are mapped out. An approximate opacity value was given to each of them based on the ratio of accessible/visible area and total area of the building. Camden council office building and Holborn Police Station are selected as political entities who operates the CCTV system. Town Hall as a democratic service place is where the public can attend council meetings to make their decision accountable. A restaurant as a private property but open to public is chosen as a semi-public space. CCTV inside is operated by its owner and staff. Euston Station is chosen as a typical public service infrastructure where CCTV's density is very high. The residents living in the borough pay council tax to the council so they have the budget to run the system. The premise of using the services that council provides to us is to consent them watching us. However, within the council, the operator who’s literally watching isn’t the decision maker who is in position of power. As a political entity, the council is operating the system of watching but not really taking act if anything goes wrong, while the police has access to the control room and they are the one to intervene into certain incidents.
. 24 .
HOLBORN POLICE STATION
. 25 .
CAMDEN TEMPORARY TOWN HALL - CROWNDALE CENTRE
. 26 .
CAMDEN ADMINISTRATIVE OFFICE
. 27 .
KOOX RESTAURANT
. 28 .
EUSTON STATION WAITING HALL
. 29 .
. 30 .
Chapter 3 Notting Hill Carnival Facial Recognition Trial 3.1 Police’s Deployment of Facial Recognition Trial 3.2 INVESTIGATION: Interview and Scene Reconstruction 3.3 Notting Hill Carnival Operation Breakdown 3.4 Main Factors - Algorithm - Camera Feeds - Watch List - False Positive 3.5 Intrusion of Privacy - Lack of Legal Framework
Notting Hill Carnival 2016/2017
POLICE'S DEPLOYMENT OF FACIAL RECOGNITION TRIAL In August 2016, Metropolitan Police at the first time deployed Facial Recognition Trial at the Notting Hill Carnival which is the second largest Carnival in the world to celebrate African Caribbean culture in London. This operation raised heated debate as it was seen as heavy intrusion of privacy and violation of human rights. In the year 2017, a joint letter was sent to Metropolitan Police calling to scrap plans of using Facial Recognition at Notting Hill Carnival. Later in 2018, Baroness Jones, the member of House of Lord and the campaign group Big Brother Watch proposed a judicial review challenge of Metropolitan Police for using Automated Facial Recognition.
. 33 .
South Wales Police’s Deployment of Facial Recognition
South Wales Police has used automated facial recognition (AFR) technology at several events in Cardiff since the Champions League final in June 2017. They were the first force in the UK to use this technology out in the field at a large sporting event. South Wales Police’s original Home Office funding application was set against the backdrop of the upcoming UEFA Champions League Final (UCL), a week-long event taking place in May-June 2017. The event presented a spectrum of policing challenges, including large crowds and associated public order issues and terrorism risks. Following the award of Home Office funding in January 2017, South Wales Police commissioned a procurement process to select a technology partner. Several companies responded to the Invitation to Tender and NEC and its ‘NeoFace’ application was selected through a process including both product quality and value for money criterion. Following the Champions League, South Wales Police deployed the AFR Locate function at a series of events (as set out in the original project plan), including: • • • • • • •
. 34 .
Elvisfest, Porthcawl The annual Elvis Festival in Porthcawl was held on 22nd and 23rd of September 2017 Operation Fulcrum, Cardiff Operation Fulcrum was deployed on the 18th of October in Cardiff City Centre. Anthony Joshua Boxing Match, Cardiff The next major deployment of AFR was for the Anthony Joshua fight on the 28th of October 2017, held in the Principality Stadium. Autumn Rugby Internationals, Cardiff AFR was deployed during November and December at four international rugby games in Cardiff Music Concerts at Motorpoint Arena, Cardiff. In December 2017. Operation Malecite, Swansea Operation Malecite was deployed in Swansea on the 23rd December Six Nations Rugby Internationals, Cardiff AFR was deployed at three Six Nations matches in Cardiff (all of the home games): Scotland (03/02/2018), Italy (11/03/2018) and France (17/03/2017).
Metropolitan Police’s Deployment of Facial Recognition
In August 2016, Metropolitan Police Firstly deployed a Facial Recognition trial at Notting Hill Carnival with a single monitoring van. Also using NEC’s NeoFace technology, Metropolitan Police has deployed the trial six times, at: • • • •
Notting Hill Carnival in 2016 and 2017 Remembrance Day 2017 Port of Hull docks (assisting Humberside Police) in 2018 Stratford transport hub for two days in June and July 2018
According to Metropolitan Police, there will be four more Live Facial Recognition trials, ten in total, before they evaluate the trial at the end of 2018. Future deployments under consideration are: football sporting events, music festivals and transport hubs. The Met will assess and analyse the results, which will be accredited by independent authorities.
. 35 .
Interview with Griff Ferris
INVESTIGATION Sketch from Interviewee
An interview was conducted with Griff Ferris who is working for campaign group Big Brother Watch and was invited to supervise the facial recognition trial deployed at Notting Hill 2017. From the interview, deployment scene was reconstructed in 3D.
. 36 .
Interview with Griff Ferris
INVESTIGATION Transcript of Interview
. 37 .
Scene Reconstruction form Interview
INVESTIGATION
Location : "... it was there, that junction, looking towards to the most footfall way from underground station"
Plan View : "... there is a van with cameras on top, it was surrounded by corrugated iron"
. 38 .
Scene Reconstruction form Interview
INVESTIGATION
Control Room : two operators inside the van with a bench of screens showing camera feeds and matches
Stop and Check: alert is sent to the police officer near the junction to go up and check the person
. 39 .
FACIAL RECOGNITION TRIAL
. 40 .
NOTTING HILL CARNIVAL OPERATION BREAKDOWN
. 41 .
Main Factors
Facial Recognition Algorithm
A Japanese company called NEC was contracted to provide the Facial Recognition platform in this trial. As research shows that facial recognition algorithm might have racial bias when it comes to identify different ethnic groups. NEC also provided hardware suggestions and trainings to Metropolitan Police operators.
NEC
Suggesting Hardware Specs and Manual Configurations
Metropolitan Police
Technician Training Workshop Police Operators
Traning Datasets
Untrained Machine Leraning Model
Training Model
. 42 .
User Interface
Packaged Algorithm for Police Operation
Main Factors
Camera Feeds
The camera on top of the van are two pan, rotate and tilt cameras. The camera feeds are manually configured. This event and location were chosen by Metropolitan Police.
Manual Configuring Settings
Camera Live Feed
Metropolitan Police Rotate, Tilt and Zoom CCTV Camera
Police Operator Decide Location and Time Notting Hill Carnival
Mobile Control Room
Crime Rates
Crowd Control
. 43 .
Main Factors
Watch List
Watch list contains targeted individuals who will be flagged out through the system when they walk into the camera viewing area. The watch list is generated from National Police Database, which contains both convicted people and unconvicted people. Despite a high court ruling in 2012 that it was unlawful to retain innocent people's images, those unconvictted people's images are not deleted yet since the central database doesn't talk to local ones automatically. The criteria of watch list selection remains unclear.
Select Detained People after Arrest
Take Photos of Detained People
Custody Images Watch List
National Police Databaase
Convicted People
Unconvicted People Manually Delet
Request to Delete New Platform to be Built . 44 .
Main Factors
False Positive
The threshold is a manually configured value. If the similarity of the person scanned lively match to the database photos reached the threshold value, the system would generate an alert. The police operator sitting inside the van will then assess if the alert is a true match, then choose to send it to police officers outside near the view area or dismiss the alert if it is a false positive. It was reported that during these two days trial, 98% alerts generated by the system was false positive. The threshold value was recommended by NEC to be set as 0.55 at the beginning, but was then adjusted to 0.59 to reduce false positive.
Facial Recognition Algorithm
Manual Configuration
Threshold Bar NeoFace Watch
Match above Threshold
Assessed Generate Alarm
Camera Feeds
Send Alarm
Police Operator A
Stop and Search Police Officer B outside of Control Room Identify Identify
Yes Arrest
Watch List
No Release
Footage Being Stored for 30 Days
. 45 .
. 46 .
Lack of Legal Framework
INTRUSION OF PRIVACY 14/01/2019
Article 8 of the European Convention on Human Rights - Wikipedia
Article 8 of the European Convention on Human Rights Article 8 of the European Convention on Human Rights provides a right to respect for one's "private and family life, his home and his correspondence", subject to certain restrictions that are "in accordance with law" and "necessary in a democratic society". The European Convention on Human Rights (ECHR) (formally the Convention for the Protection of Human Rights and Fundamental Freedoms) is an international treaty to protect human rights and fundamental freedoms in Europe.
Contents Right Private life Family life Home
Breakdown of the Notting Hill Carnival Facial Recognition Tri-
Case law al shows that many processes in different aspects can contribCases involving LGBT rights
ute to its discriminatory results. The Facial Recognition Trial is intrusive, because it uses people's facial images without them knowing it. The whole operation was conducted inside a van surrounded by corrugated iron, which is neither acknowledging the public what is happening nor asking people's consent if they are willing to be scanned during the trial.
Violation of the convention by mass surveillance See also Notes External links
Right
There is still lack of regulation and legal frame specifically about facial recognition. Biometric Strategy issued by Home Office in Article 8 – Right to respect for private and family life June 2018 failed to address the issue clearly.
1. Everyone has the right to respect for his private and family life, his home and his correspondence. 2. There shall be no interference by a public authority with the exercise of this right except such as is in accordance with the law and is necessary in a democratic society in the interests of national security, public safety or the economic well-being of the country, for the prevention of disorder or crime, for the protection of health or morals, or for the protection of the rights and freedoms of others. Article 8 is considered to be one of the Convention's most open-ended provisions.[1]
Private life For better understanding of perception of "private life" case law should be analyzed. In Niemietz v. Germany, the Court held that it "does not consider it possible or necessary to attempt an exhaustive definition of the notion of 'private life'. However, it would be too restrictive to limit the notion to an 'inner circle' in which the individual may live his own personal life as he choose and to exclude therefrom entirely the outside world not encompassed within that circle. Respect for private life must also comprise[2] to a certain degree the right to establish and develop relationship and develop relationship with other human beings."
https://en.wikipedia.org/wiki/Article_8_of_the_European_Convention_on_Human_Rights
1/4
. 47 .
. 48 .
Chapter 4 CCTV 4.1 Evolution of CCTV 4.2 Public CCTV - Conduct Code 4.3 Private CCTV - Conduct Code 4.4 Types of CCTV Camera 4.5 TEST 01: Specs Affecting Capturing Space - Focal Length & Resolution - Visualising View Cone 4.6 Data Storage and Retention 4.7 TEST 02: Specs Affecting Capturing Activities - Frame Rate & Resolution - Mapping Activities 4.8 Data Retention Protocol 4.9 Control Room Design 4.10 View from Control Room - Fragmented City 4.11 TEST 03: View from Control Room - CCTV View Range in Euston Station - Interior Room CCTV Analogue - Photogrammetry Test
EVOLUTION OF CCTV
1940
1950
The earliest documented use of CCTV technology was in Germany in 1942. The system was designed by the engineer Walter Bruch and it was set up for the monitoring of V-2 rockets. It wasn’t until 1949 that the technology was launched on a commercial basis. In that year, an American government contractor named Vericon began promoting the system.
1960
1970
A major development in the history of CCTV occurred when video cassette recordings (VCRs) became widely available in the 1970s First CCTV System available
1980
1990
2000
Multiplexer Time-lapse systems become widespread 1985 – After many unsuccessful trials in the 1970’s, the UK took a leap of faith and installed the first outdoor CCTV system in the beautiful seaside town, Bournemouth. 1987 – In King’s Lynn, Norfolk, the council set up the first local government surveillance system. The installing of the cameras were proven successful in deterring crime. This resulted in a dramatic increase in CCTV cameras being installed in public places. Colour CCTV cameras hit mainstream First IP Cameras begin to appear on market Digital Video Recorders (DVR) dominate the industry IP based video systems sppear, dubbed NVR ‘Network Video Recorders’ IP Video becomes cost-effective for new and existing installations
2010
2013 – The British Security Industry Authority (BSIA) estimated that Britain has a CCTV camera for every 11 people. That’s a whopping 5.9 million active cameras in the country.
2020
. 51 .
Surveillance Camera Code of Practice
PUBLIC CCTV CAMERAS
The role of Surveillance Camera Commissioner (SCC) was created under the Protection of Freedoms Act 2012 (PoFA). The SCC was appointed by the Home Secretary and is independent of government. The SCC’s statutory functions are to encourage compliance with the Home Secretary’s Surveillance Camera Code of Practice (the SC Code) and its 12 guiding principles, which if followed will ensure that surveillance camera systems are only operated proportionately, transparently and effectively. The SC Code applies to the overt use of surveillance camera systems that are operated by relevant authorities only (police forces, local authorities and parish councils) in public places in England and Wales, regardless of whether or not there is any live viewing or recording of images or information or associated data. (Home Office: https://www.gov.uk/government/publications/ domestic-cctv-using-cctv-systems-on-your-property/domesticcctv-using-cctv-systems-on-your-property)
. 52 .
PUBLIC CCTV CAMERAS
. 53 .
Surveillance Camera Code of Practice
PRIVATE CCTV CAMERAS
An individual has the right to protect their property and this can be done by using a CCTV system where it is necessary, such as a security measure. However, the Surveillance Camera Commissioner (SCC) recommends that users of CCTV systems should operate them in a responsible way to respect the privacy of others. You also need to be aware that if your camera captures images outside the boundaries of your household, then the GDPR and DPA will apply to you, and you will need to ensure your use of CCTV complies with these laws. Ensure that you are transparent to those around you when installing your CCTV system. You can do this by: • •
informing your neighbour(s) about your system putting up a notice informing people that recording is taking place
(Home Office: https://www.gov.uk/government/publications/ domestic-cctv-using-cctv-systems-on-your-property/domesticcctv-using-cctv-systems-on-your-property)
. 54 .
PRIVATE CCTV CAMERAS
. 55 .
Surveillance Camera Code of Practice
TYPES OF CCTV CAMERA
. 56 .
Dome Camera The camera of this CCTV is eyeball shaped. Dome cameras are commonly used for indoor security and surveillance. The shape of the camera makes it difficult for onlookers to tell which way the camera is facing, which is a strong piece of design, deterring criminals by creating an air of uncertainty. Easy to install – It only requires two or more screws to install a dome camera. It can easily be mounted on both vertical and horizontal areas. Vandal-proof feature – The dome-shaped casing covers the camera and protects it from vandalism. Infrared capable – The dome camera is fitted with IR illuminators which enables it to capture video images, even in low light conditions.
Bullet Camera This CCTV is a long, cylindrical and tapered shape camera that looks like a rifle bullet.Their strengths lie specifically in applications which require long distance viewing. Installed within protective casings, the cameras are protected against dust, dirt and other natural elements. The cameras can easily be mounted with a mounting bracket, and come fitted with either fixed or varifocal lenses depending on the requirements of its intended application. Other benefits of bullet cameras include: Adaptability – can use indoors and outdoors Compact size aids installation High quality image resolution
C-mount Camera Coming with detachable lenses, C-mount cameras allow for simple lens changes to fit different applications. C-mount cameras can cover distances beyond 40 ft thanks to the possibility to use special lenses with these cameras, whereas standard CCTV lenses can only cover distances of 35-40 ft. Other benefits of C-mount cameras include: Can support changes in technology Effective for indoor use Bulky size makes them noticeable (which acts as a deterrent)
Day/Night Camera Capable of operating in both normal and poorly lit environments, these cameras benefit from not requiring inbuilt infrared illuminators as they can capture clear video images in the dark thanks to their extra sensitive imaging chips. For this reason, these cameras are ideal for outdoor surveillance applications in which IR cameras are unable to function optimally. Other benefits of day/night cameras include: Record in both colour and black & white. Wide variety of sizes available Infrared capability
PTZ Pan Tilt & Zoom Camera PTZ – Pan/tilt/zoom – cameras allow the camera to be moved left or right (panning), up and down (tilting) and even allow the lens to be zoomed closer or farther. These cameras are used in situations where a live guard or surveillance specialist is there operating the security systems. Other benefits of PTZ cameras include: 200m IR night vision X36 optical zoom High-quality image resolution
Network/IP CCTV Camera These cameras share the images across the internet, so CCTV footage can be easily accessed. Network cameras are ideal for both domestic and commercial purposes because you can see what’s going on whilst away from the property. Other benefits of network cameras include: Data can be easily accessed Ideal for homes and companies Less cabling and less maintenance . 57 .
Test 01: Specs Affecting Capturing Space
FOCAL LENGTH AND RESOLUTION
Lens
2.8 mm
3.6 mm
6 mm
8 mm
12 mm
16 mm
Angle
90o
69o
50o
33o
22o
18o
Distance
3-5m
8m
15 m
20 - 30 m
40 - 50 m
70 - 80 m
Image Sensor
Lens
5m
8m
2.8 mm
3.6 mm
15 m
40 m
Focal Length
. 58 .
6 mm
8 mm
12 mm
Test 01: Specs Affecting Capturing Space
TOOL KIT
Generally, the CCTV camera lenses are much wider than normal camera lens, which makes it difficult to simulate the CCTV lens view with normal camera. In order to achieve the effect, a wide angle lens kit was used in this case. Within this kit, Wide Angle Lens (0.36&0.63X) were equipped on iPhone X camera, with iPhone camera's inbuilt zoom in x2.0. The test compared fourth different focal length with five different resolution to demonstrate how camera hardware configuration could affect space capturing. Test video were taken at the same spot where Notting Hill Carnival Facial Recognition Van was located.
. 59 .
Test 01: Specs Affecting Capturing Space
FOCAL LENGTH AND RESOLUTION ×0.36
Full HD 1920×1080 px
Standard HD 1280×720 px
D1 720×480 px
CIF 352×240 px
QCIF 176×120 px . 60 .
×0.63
Test 01: Specs Affecting Capturing Space
FOCAL LENGTH AND RESOLUTION ×1.0
×2.0
Full HD 1920×1080 px
Standard HD 1280×720 px
D1 720×480 px
CIF 352×240 px
QCIF 176×120 px . 61 .
Test 01: Specs Affecting Capturing Space
VIEW CONE OF WIDE ANGLE x 0.36 ×0.36
Full HD 1920×1080 px
Standard HD 1280×720 px
D1 720×480 px
CIF 352×240 px
QCIF 176×120 px
. 62 .
Test 01: Specs Affecting Capturing Space
VIEW CONE OF WIDE ANGLE x 0.63 ×0.63
Full HD 1920×1080 px
Standard HD 1280×720 px
D1 720×480 px
CIF 352×240 px
QCIF 176×120 px
. 63 .
Test 01: Specs Affecting Capturing Space
VIEW CONE OF IPHONE X DEFAULT LENS x 1.0 ×1.0
Full HD 1920×1080 px
Standard HD 1280×720 px
D1 720×480 px
CIF 352×240 px
QCIF 176×120 px
. 64 .
Test 01: Specs Affecting Capturing Space
VIEW CONE OF IPHONE X DEFAULT LENS x 2.0 ×2.0
Full HD 1920×1080 px
Standard HD 1280×720 px
D1 720×480 px
CIF 352×240 px
QCIF 176×120 px
. 65 .
Test 01: Specs Affecting Capturing Space
FOCAL LENGTH AND RESOLUTION
. 66 .
From this test, it is evident that wider length can capture more space. With the same resolution, wider length see closer subjects while narrower length can see subject further clearer.
. 67 .
Data Storage and Retention DATA STORAGE AND RETENTION Development of Data Storage Technology Traditional CCTV camera system: with VCR CCTV systems relay on analog signal transmission, it uses dedicated cameras, cables, multiplexers, video recorders and monitors. The images from the analog video surveillance camera, CCTV which mounted in a fixed location are transmitted via a copper coaxial cable for viewing on a dedicated monitor and recording on the video cassette recorder (VCR). A VCR, or video recorder is an electromechanical device that records analog audio and analog video from broadcast television or other source on a removable, magnetic tape videocassette, and can play back the recording.
(Wh-tech.com, 2010)
DVR connected directly to analog CCTV camera, devoted to digitize and compress the video images and store them onto the hard disk. DVR technology is in the process of replacing the traditional VCR technology used in many CCTV camera systems. A DVR provides better recording image quality and longer recording duration than VCR, you can also retrieving video images from any specific time or date of CCTV camera.
(Wh-tech.com, 2010) . 68 .
Most CCTV systems may record and store digital video and images to a digital video recorder (DVR) or, in the case of IP cameras, directly to a server, either on-site or off-site. There is a cost in the retention of the images produced by CCTV systems. The amount and quality of data stored on storage media is subject to compression ratios, images stored per second, image size and is effected by the retention period of the videos or images. DVRs store images in a variety of proprietary file formats. Recordings may be retained for a pre-set amount of time and then automatically archived, overwritten or deleted, the period being determined by the organisation that generated them.
Network IP-Surveillance by Video server, are the key to transforming analog video into digital video, distribute compressed live video over a LAN or Internet, making it possible to migrate toward a digital system without having to discard functional analog equipment. Video servers uses standard PC server for video recording and management, connect your existing analog CCTV camera to video server and connect the video server to the IP network and the video images can be transmitted to any PCs in any locations and be stored on computer hard drives.
(Wh-tech.com, 2010)
IP-Surveillance system with network cameras. IP-Surveillance connect the Ethernet cable to the network, the network camera can work independently and do not need to be connected to any PCs, you can operate the camera, view and record the video images at the remote location wherever there is an IP network.
(Wh-tech.com, 2010)
. 69 .
Test 02: Specs Affecting Capturing Activities
FRAME RATE AND RESOLUTION 30 fps
15 fps
7.5 fps
5 fps
1 fps
QCIF 176×120 px
. 70 .
CIF 352×240 px
D1 720×480 px
Standard HD 12
280×720 px
Frame rate (expressed in frames per second or fps) is the frequency (rate) at which consecutive images called frames appear on a display. Frame rate and resolution are two main factors affecting video storage. Resolution relates to the size of each frame, and frame rate relates to quantity of the frames recorded. In order to see subjects clearly, the focal length setting of x2 was selected in this test. Frame rate of 30, 15, 7.5 , 5 and 1 were tested alongside with 5 different resolution settings. Frame rate can affect how much activities of the subject can be recoded. Test video were taken at the same spot where Notting Hill Carnival Facial Recognition Van was located.
Full HD 1920×1080 px
. 71 .
. 72 .
Test 02: Specs Affecting Capturing Activities
30 FPS
. 73 .
. 74 .
Test 02: Specs Affecting Capturing Activities
15 FPS
. 75 .
. 76 .
Test 02: Specs Affecting Capturing Activities
7.5 FPS
. 77 .
. 78 .
Test 02: Specs Affecting Capturing Activities
5 FPS
. 79 .
. 80 .
Test 02: Specs Affecting Capturing Activities
1 FPS
. 81 .
Test 02: Specs Affecting Capturing Activities
FRAME RATE AND RESOLUTION
. 82 .
30 fps
15 fps
5 fps
1 fps
7.5 fps
Five recorded subjects' activities were mapped frame by frame for ten seconds. From the mapping, it is evident that 1 fps is already enough to track subjects movement within captured space. Normally, CCTV's frame rate is set to be 7.5 when subjects' movements are not too rapid. 15 fps can record activities in detail. In the case of Notting Hill Carnival, crowds are moving slowly and many people are captured in CCTV's view, the frame rate was set to be 10 fps in the end.
. 83 .
(Statewatch.org, 2007)
. 84 .
DATA RETENTION PROTOCOL
The amount of time that security video data needs to be retained can be an important factor in many issues. This data needs to be preserved for a range of purposes, from the primary reason it was created (such as monitoring a food processing line) as well as for a reasonable amount of time to recover any evidence of any other important activity that it might document (such as a contaminant entering the food processing line). Also, this data may need to be reviewed for any historical, research or other long-term information of value it may contain, such as for improving supply chain management or manufacturing processes. According to the National CCTV Strategy report from the Home Office in the U.K., it has long been accepted that CCTV recordings should routinely be kept between 28 and 31 days before being recorded over. This time period allows the police the opportunity to recover CCTV evidence and respond to lines of enquiry that may not have been known at the time an incident was reported. This schedule also helped in the videotape recycling process by ensuring that the tape would not be used day in and day out. With the introduction of digital CCTV systems, some system owners have moved away from the 28 and 31 days figure to periods as short as 14 days. However, this has resulted in significant resource implications for the police, which must collect the digital CCTV before the footage is overwritten. This becomes even more important in terrorist investigations, where extended periods of between 14 and 31 days are often required. With that in mind, the police have therefore reiterated their need for 31 days of storage in the digital CCTV era, with the requirement that the recording quality should not be reduced or compromised, which could result in the recordings not meeting the “fit for purpose” criterion. On some occasions, such as cold cases and trials undergoing appeals, CCTV material has to be archived for a number of years.
. 85 .
Control Room Design
PROJECT CYBERSYN
https://99percentinvisible.org/episode/project-cybersyn/ (Mingle, 2016)
Project Cybersyn was a Chilean project from 1971–1973 during the presidency of Salvador Allende aimed at constructing a distributed decision support system to aid in the management of the national economy. The principal architect of the system was British operations research scientist Stafford Beer, and the system embodied his notions of organisational cybernetics in industrial management. One of its main objectives was to devolve decision-making power within industrial enterprises to their workforce in order to develop self-regulation of factories.
. 86 .
Sketch Perspective of the Operation Room
Sketch Plan of the Operation Room
Tulip Chair
Plan of the Operation Room
Photo of the Operation Room
https://99percentinvisible.org/episode/project-cybersyn/ (Mingle, 2016)
Regardless of what aim Cybersyn Project system was trying to achieve, the design of its control room gives us an inchoate understanding of contemporary CCTV control room. Its hexagonal layout maximised every seat’s view towards all the monitors. Each Tulip chair ad an ashtray, a place for whiskey glass and a set of buttons that controlled the display screen on the walls.
. 87 .
Control Room Design
CONTEMPORARY CONTROL ROOM
https://www.2020cctv.com/security-solutions/cctv-command-control/ (2020 Vision Systems, 2019)
CCTV Control Room or operations room is a room serving as a central space where a large physical facility or physically dispersed service can be monitored and controlled. A control room will often be part of a larger command center. Control rooms for vital facilities are typically tightly secured and inaccessible to the general public. Multiple electronic displays and control panels are usually present, and there may also be a large wall-sized display area visible from all locations within the space. Some control rooms are themselves under continuous video surveillance and recording, for security and personnel accountability purposes. Many control rooms are manned on a “24/7/365” basis, and may have multiple people on duty at all times (such as implementation of a “two-man rule”), to ensure continuous vigilance.
. 88 .
Size
Space needed for control room has always been underestimated as workstations, monitors and other equipments would take a lot of areas. Roof heights, office pillars and walls should all be taken into consideration while the control room was designed.
Light
Light can be a factor which can make or break the effectiveness of a surveillance control room. Big open windows and bright overhead lighting should be avoided in control rooms, as sunlight can reflect off monitors, and bright strip lights can cause eye fatigue at night. Lighting levels in general should be much lower than typical office lighting. Controlled natural light is useful as its typically a more subtle and less irritating light source for operators. Any windows should have easy to control blinds, with frosting considered depending which direction they face.
Noise Control Modern control rooms don’t just watch from afar, they interact with staff, personnel and the public; and so it’s important that noise levels are low enough to allow clear communication when it’s needed. It also allows them to concentrate on what can be the taxing job of juggling multiple systems and tasks at the same time. Therefore, the location of the control room should be taken into consideration to avoid noises, and good level of sound insulation should be provided if it is close to source of noise pollution.
Sights
Surveillance control rooms rely on reliable, quick-reacting personnel, who should be able to seamlessly work together and see what they need to see at all times. Personnel in surveillance control rooms may find themselves looking up at large monitors regularly, so it’s essential that important line of sights are clear.
Work Interfaces Usable desk space is also highly important, as personnel in control rooms will have to frequently make notes in logbooks and other documentation while observing monitors in front of them. When implementing double or triple monitor set ups on personnel desks, ensure that they are comfortably in the personnels vision. As for wall mounted monitor displays, ensure that they are positioned and angled correctly so that the risk of glare and reflection can be reduced.
. 89 .
Fragmented and Flattened City
VIEW FROM CONTROL ROOM
Islington CCTV Control Room (Draper, 2018)
. 90 .
. 91 .
Test 03: View from Control Room
CCTV VIEW RANGE IN EUSTON STATION
. 92 .
Test 03: View from Control Room
INTERIOR PHOTOGRAMMETRY CAMERA SETUP
In the case of Euston Station's waiting hall, CCTV cameras surround the whole space from the ceiling wich ensures their view range can cover as much space as possible. The principle of setting cameras for photogrammetry is similar to setting CCTV for interior space as the diagram shown above. Therefore, photgrammetry an interior room can be used as a simulation to potrary a space from CCTV control room's perspective.
. 93 .
Test 03: View from Control Room
PHOTOGRAMMETRY TEST
. 94 .
. 95 .
Test 03: View from Control Room
PHOTOGRAMMETRY TEST
. 96 .
Although cameras were set to capture a space from all possible perspectives, there are still blink spots which cannot be captured. So as CCTVs dispersed in urban space. The view inside control room is fragmented with some overlapping and a lot of unseen beyond frames. . 97 .
. 98 .
Chapter 5 Facial Recognition 5.1 Facial Recognition Overview 5.2 Deep Neutral Network 5.3 Facial Recognition Framework 5.4 Applications of Facial Recognition 5.5 Principles and Coding Structure 5.6 TEST 04: Fabricating Identities - Database Structure - Real Identities - Exchange Identities - Fake Identities 5.7 TEST 05: Architectural Elements Affecting Facial Recognition - Distance and Perspective - Lighting Level - Blockage / Barriers - Material - Translucency - Reflection
FACIAL RECOGNITION OVERVIEW 1. Finding a Face Systems extract patterns from an image and compare them to a model of a face. When patterns start to resemble the model, the system signals it has homed in on a face.
Personal devices
Checkpoint cameras
Other cameras
CCTV cameras
Smartphones use face recognition for apps and security, such as unlocking the phone.
Faces are recorded at customs and security checkpoints, and the images are archived.
Laptop, video, and thermal cameras used in some security systems can capture face images.
Systems can isolate and track individuals by face, gait, and clothing color and pattern.
2. Creating a face template
Face imagery captured when a person poses for the camera, such as at security checkpoints, is easier to analyze; imagery captured from CCTV cameras may require advanced methods and detailed analysis.
Algorithms build more informative and accurate digital representations—called face templates—using thermal, geometric, and other data, either separately or combined.
Geometric Spatial relationships between facial features, such as the center of the eyes and tip of the nose, are calculated.
Photometric
Skin-texture analysis
Algorithms can build a Pores, wrinkles, and spots face even if an image is are mapped and analyzed; obscured by poor lighting the technology can even or distorted by odd an- differentiate between gles or expressions. twins.
Thermal sensors This technology can provide further information despite obstacles such as heavy makeup or disguises.
2. Creating a face template Algorithms build more informative and accurate digital representations—called face templates—using thermal, geometric, and other data, either separately or combined.
Identity confirmed
A facial recognition system is a technology capable of identifying or verifying a person from a digital image or a video frame from a video source. There are multiple methods in which facial recognition systems work, but in general, they work by comparing selected facial features from given image with faces within a database. It is also described as a Biometric Artificial Intelligence based application that can uniquely identify a person by analysing patterns based on the person’s facial textures and shape. While initially a form of computer application, it has seen wider uses in recent times on mobile platforms and in other forms of technology, such as robotics. It is typically used as access control in security systems and can be compared to other biometrics such as fingerprint or eye iris recognition systems. Although the accuracy of facial recognition system as a biometric technology is lower than iris recognition and fingerprint recognition, it is widely adopted due to its contactless and non-invasive process. Recently, it has also become popular as a commercial identification and marketing tool. Other applications include advanced human-computer interaction, video surveillance, automatic indexing of images, and video database, among others. . 101 .
Deep Neural Network
HUMAN NEURON Dendrite
Nucleus
Soma Axon
Myelin sheath
Schwann cell
Axon Terminal Electrical Signal
Synapse
An artificial neuron is a mathematical function conceived as a model of biological neurons, a neural network. Artificial neurons are elementary units in an artificial neural network. The artificial neuron receives one or more inputs (representing excitatory postsynaptic potentials and inhibitory postsynaptic potentials at neural dendrites) and sums them to produce an output (or activation, representing a neuron’s action potential which is transmitted along its axon). Usually each input is separately weighted, and the sum is passed through a non-linear function known as an activation function or transfer function. The transfer functions usually have a sigmoid shape, but they may also take the form of other non-linear functions, piecewise linear functions, or step functions. They are also often monotonically increasing, continuous, differentiable and bounded. The thresholding function has inspired building logic gates referred to as threshold logic; applicable to building logic circuits resembling brain processing. For example, new devices such as memristors have been extensively used to develop such logic in recent times. . 102 .
Deep Neural Network
ARTIFICIAL NEURON
Inputs
Weights
X1
Sum Activation Fuction
w1 X2
w2
X3
∑ f(x)
w3 w4
X4 . . .
Output
f(∑w x ) i
i
wn
Xn
Data Processing
Input Layer
Hidden Layer 1
Hidden Layer 2
Hidden Layer 3
X1
Output Layer
X2 X3 X4 X5 X6 X. 7 . .
Xn
Synapse
Synapse
Synapse
Synapse
Deep Neural Architecture with Multiple Layers . 103 .
Review of Face Detection Systems
FACIAL RECOGNITION SYSTEM
(AL-Allaf, 2014)
A general face recognition system includes many steps: face detection; feature extraction; and face recognition. Face detection and recognition includes many complementary parts, each part is a complement to the other. Depending on regular system each part can work individually. Face detection is a computer technology that is based on learning algorithms to allocate human faces in digital images. Face detection takes images/video sequences as input and locates face areas within these images. This is done by separating face areas from non-face background regions. Facial feature extraction locates important feature (eyes, mouth, nose and eye-brows) positions within a detected face. Feature extraction simplifies face region normalization where detected face aligned to coordinate framework to reduce the large variances introduced by different face scales and poses. The accurate locations of feature points sampling the shape of facial features provide input parameters for the face identification. Other face analysis task: facial expression analysis; face animation and face synthesis can be simplified by accurate localization of facial features. Face identification generates the final output of complete face-recognition system: the identity of the given face image. Based on normalized face image and facial feature locations derived from previous stages, a feature vector is generated from given face and compared with a database of known faces. If a close match is found, the algorithm returns the associated identity. A main problem in face identification is the large differences between face images from the same person as compared to those from different persons. Therefore, it is important to choose a suitable face classification technique that can provide a good separate ability between different persons. Face identification has a wide range of applications. Because it offers a non-intrusive way for human identification, the face is used as an important biometric in security applications.
. 104 .
Retinal Connected Neural Network (RCNN)
(AL-Allaf, 2014) Rowley, Baluja and Kanade (1996) presented face detection system based on a retinal connected neural network (RCNN) that examine small windows of an image to decide whether each window contains a face. The system arbitrates between many networks to improve performance over one network. They used a bootstrap algorithm as training progresses for training networks to add false detections into the training set. This eliminates the difficult task of manually selecting non-face training examples, which must be chosen to span the entire space of non-face images.
Rotation Invariant Neural Network (RINN)
(AL-Allaf, 2014)
Rowley, Baluja and Kanade (1997) presented a neural network-based face detection system. Unlike similar systems which are limited to detecting upright, frontal faces, this system detects faces at any degree of rotation in the image plane. The system employs multiple networks; the first is a “router” network which processes each input window to determine its orientation and then uses this information to prepare the window for one or more detector networks.
. 105 .
Principal Component Analysis with ANN (PCA & ANN)
(AL-Allaf, 2014) Jeffrey Norris (1999) used using principal component analysis (PCA) with class specific linear projection to detect and recognized faces in a real-time video stream. The system sends commands to an automatic sliding door, speech synthesizer, and touchscreen through a multi-client door control server. Matlab, C, and Java were used for developing system. The system steps to search for a face in an image: 1. Select every 20×20 region of input image. 2. Use intensity values of its pixels as 400 inputs to ANN. 3. Feed values forward through ANN, and 4. If the value is above 0.5, the region represents a face. 5. Repeat steps (1..4) several times, each time on a resized version of the original input image to search for faces at different scales.
Convolutional Neural Network (CNN)
(AL-Allaf, 2014) Masakazu Matsugu (2003) described a rule-based algorithm for robust facial expression recognition combined with face detection using a convolutional neural network (CNN).
. 106 .
Evolutionary Optimization of Neural Networks
(AL-Allaf, 2014) Stefan, et al (2004) used ANN to get decision whether a pre-processed image region represents a human face or not. They described the optimization of this network by a hybrid algorithm combining evolutionary computation and gradient-based learning. The evolved solutions perform considerably faster than an expert-designed architecture without loss of accuracy. The proposed hybrid algorithm tackles the problem of reducing the number of hidden neurons of face detection network without loss of detection accuracy. The speed of classification whether an image region corresponds to a face or not could be improved by approximately 30 %.
Evolutionary Optimization of Neural Networks
(AL-Allaf, 2014) According to Rowley work in, Marian Beszedes & Milos Oravec (2005) presented a neural network based face detection system to detect faces in an unprocessed input image. They used image processing techniques such as normalization, rotation and position, light conditions improvement on small windows extracted from the input image. Multilayer perceptron (MLP) used to detect rotation of input window and also to decide whether the window contains a face or not. This system is based on method to distribute the decision among multiple sub networks and an algorithm is used to train this ANN and the result of this system is in the form of a set containing locations of human faces.
. 107 .
APPLICATIONS OF FACIAL RECOGNITION
. 108 .
Apple introduced Face ID Apple introduced Face ID on the flagship iPhone X as a biometric authentication successor to the Touch ID, a fingerprint based system. Face ID has a facial recognition sensor that consists of two parts: a “Romeo” module that projects more than 30,000 infrared dots onto the user’s face, and a “Juliet” module that reads the pattern. The pattern is sent to a local “Secure Enclave” in the device’s central processing unit (CPU) to confirm a match with the phone owner’s face. The facial pattern is not accessible by Apple. The system will not work with eyes closed, in an effort to prevent unauthorized access. The technology learns from changes in a user’s appearance, and therefore works with hats, scarves, glasses, and many sunglasses, beard and makeup. It also works in the dark. This is done by using a “Flood Illuminator”, which is a dedicated infrared flash that throws out invisible infrared light onto the user’s face to properly read the 30,000 facial points.
Eurostar implements facial recognition for Paris passengers Launched in cooperation with Eurostar, the French border police and the French Ministry of Interior, the ABC equipment, created by Vision-Box, will make the immigration procedure for those travelling from the UK into Europe much faster and smoother. This also marks both the first time that self-service technology has been installed by a private transport service and that such a programme has been used at a French border. Passengers with biometric passports scan the document at a first gate and then proceed to a second gate where a camera compares their features with their passport picture before allowing them through to the departure lounge.
Facial recognition software helps diagnose rare genetic disease Researchers with the National Human Genome Research Institute (NHGRI), part of the National Institutes of Health, and their collaborators, have successfully used facial recognition software to diagnose a rare, genetic disease in Africans, Asians and Latin Americans. The disease, 22q11.2 deletion syndrome, also known as DiGeorge syndrome and velocardiofacial syndrome, affects from 1 in 3,000 to 1 in 6,000 children. Because the disease results in multiple defects throughout the body, including cleft palate, heart defects, a characteristic facial appearance and learning problems, healthcare providers often can’t pinpoint the disease, especially in diverse populations.
Facial recognition used in shops' slef-checkout machine Facial recognition will soon be used in British shops for the first time to judge how old customers are at self-checkout machines when they buy age restricted items, it is understood following a deal with the company that makes the tills for Tesco and Asda. NCR, who make the software and hardware for self check machines for the majority of the UK’s supermarkets, will integrate a camera that will estimate the age of shoppers when they are buying alcohol and cigarettes. It will mean a member of staff will not have to intervene to approve a purchase, but instead the machine will scan a person’s face and either accept or deny the sale of the item, without needing to see any official identification from the customer. The system does not require shoppers to register their identity in advance and does not retain any visual information about users after they have made a purchase. . 109 .
PRINCIPLES Facial Recognition algorithm used for test in this chapter is based on open sourced python codes and facial detection library. Built using dlib’s state-of-the-art face recognition built with deep learning. The model has an accuracy of 99.38% on the Labeled Faces in the Wild benchmark.
Step 1: Finding all the Faces
Step 2: Posing and Projecting Faces
Step 3: Encoding Faces
Step 4: Finding the person’s name from the encoding
. 110 .
CODING STRUCTURE
Facial Recognition GUI Kivy library is used to implement the graphical user interface for the facial recognition application.
Facial Recognition Dlib library and face_recognition api were used to implement the algorithm of face recognition
np array
Files to Database This module interprets the original personal text information and image profile into a np array. facial recognition module's function is called.
np array
Stream Processor Firstly return the current frame of the Webcam stream as np array, then analyse the stream and return in real time the name, distance and location of people in the video stream.
LogFileWriter A class to keep logs of different actions realised
Facial recognition coding was implemented with Python in Spyder with open sourced codes from Github. https://github.com/arupiot/aiml_eye/tree/master/facial_recognition_ manager contributed by Francesco Anselmo. OpenCV, Dilib and Kivy are main libraries used in this algorithm. Also used functions from face_recognition api (https://github.com/ageitgey/ face_recognition)
. 111 .
Python
CODES
. 112 .
. 113 .
Test 04: Fabricating Identities
DATABASE STRUCTURE
. 114 .
. 115 .
Test 04: Fabricating Identities
REAL IDENTITIES CARD: Linn BIO: She is an architectural student at the AA
CARD: Zhenyan BIO: She is a composer from RAM
. 116 .
Linn and Zhenyan are recognised correctly with bio cards information matching their profile photos. In this test, different identities were given to Linn and Zhenyan by manipulating their Bio Cards. It is evident that the database (watch list) for facial recognition is a manual configuration which can be completely fabricated and manipulated by the system's operator.
. 117 .
Test 04: Fabricating Identities
EXCHANGE IDENTITIES
CARD: Zhenyan BIO: She is a composer from RAM
CARD: Linn BIO: She is an architectural student at the AA
When reversed identity bio cards were created, the algorithm recognised Linn and Zhenyan in an opposite way.
. 118 .
Test 04: Fabricating Identities
FAKE IDENTITIES
CARD: Anna BIO: She is an architectural student at the AA
CARD: Li BIO: She is a composer from RAM
When fake identity bio cards were created, the algorithm recognised Linn and Zhenyan with faked identity (Anna and Li respectively).
. 119 .
Test 05: Architectural Elements Affecting Facial Recognition
SPATIAL EFFECTS TESTS SETTING
. 120 .
Facial recognition algorithm was run on a desktop with video stream captured by a HD Webcam (specs shown on left).
. 121 .
Test 05.1: Architectural Elements Affecting Facial Recognition
DISTANCE AND PERSPECTIVE /// TEST SETUP
DISTANCE BETWEEN THE WEBCAM AND THE FACE TO BE RECOGNISED The input device for the facial recognition is the webcam in this test to simulate a CCTV camera’s view. The distance between the webcam and the face to be recognised is measured to demonstrat how far the algorithm can work to recognise a person.
. 122 .
. 123 .
200 mm
80o
60o
30o
0o
-30o
-60o
-80o . 124 .
300 mm
400 mm
500 mm
600 mm
700 mm
800 mm
900 mm
1000 mm
. 125 .
Test 05.1: Architectural Elements Affecting Facial Recognition
FACIAL RECOGNITION RANGE
. 126 .
In this test, facial recognition algorithm performs well in the distance between 400-900mm and with the perspective within 60o. When CCTV is set for facial recognition, it requires zoom into the subjects' face rather than capturing larger space with wide lens. Subjects' activity (head movement) might affect the recognition result due to the perspective of captured face. . 127 .
Test 05.2: Architectural Elements Affecting Facial Recognition
LIGHTING LEVEL
. 128 .
/// TEST SETUP
. 129 .
Test 05.2: Architectural Elements Affecting Facial Recognition
LIGHTING TEST EQUIPMENTS
Lights being used in this test are three LED lights with tripod and two diffusions of white and orange. With this setting, the brightness and the temperature of the lights can be controlled to fulfil several variations.
. 130 .
Test 05.2: Architectural Elements Affecting Facial Recognition
LIGHTING MEASURE TOOL-KIT
A light meter was used to measure lighting level in this test.
. 131 .
Test 05.2: Architectural Elements Affecting Facial Recognition
LIGHTING TEST
BRIGHTNESS / TEMPERATURE
WHITE
YELLOW
Unknown
54.37%
61.60%
64.72%
64.68%
52.82%
56.28%
51.95
61.04%
57.19%
2 lux / 6 lux
25 lux / 30 lux
50 lux
100 lux
200 lux
. 132 .
300 lux
66.83%
63.58%
64.45%
64.38%
64.22%
64.03%
62.50%
66.81%
67.85%
64.49%
500 lux
1000 lux
1500 lux
2000 lux
INTERIOR (600 lux) 68.52% . 133 .
Test 05.2: Architectural Elements Affecting Facial Recognition
LIGHTING TEST
BRIGHTNESS / TEMPERATURE
Face can be recognised successfully with even very low lighting levels. It performs better when the subject is lit evenly, not necessary to be too bright. Sharp lighting causing hard shadows may get worse recognition result. Nowadays, most advanced CCTV cameras already have night view function. Lighting as an environmental factor affecting users' spatial experience can be intentionally designed to created space more or lesss watched from CCTV's perspecrive.
. 134 .
. 135 .
Test 05.3: Architectural Elements Affecting Facial Recognition
BLOCKAGE/BARRIERS /// TEST SETUP
. 136 .
. 137 .
Test 05.3: Architectural Elements Affecting Facial Recognition
BLOCKAGE / BARRIERS TEST
. 138 .
. 139 .
Test 05.3: Architectural Elements Affecting Facial Recognition
BLOCKAGE / BARRIERS TEST As research shown in previous pages, facial recognition algorithm is recognising face based on calculation of distances between facial features. Therefore, the blockage of main facial feature would cause failure of facial recognition. In this test, the coverage of 1/3 face can still provide enough facial information for the algorithm to recognise.
. 140 .
. 141 .
Test 05.4: Architectural Elements Affecting Facial Recognition
MATERIAL - TRANSLUCENCY /// TEST SETUP
DISTANCE BETWEEN THE MATERIAL AND THE FACE TO BE RECOGNISED Translucent materials are positioned between the face and the webcam to test how the algorithm performs to recognise blurry face.
. 142 .
. 143 .
Test 05.4: Architectural Elements Affecting Facial Recognition
TRANSLUCENCY TEST - MATERIAL COLLECTION
CLEAR
Creating Translucent Materials A glass frosting paint was used to coat clear styrene sheets to create a selection of materials with different translucency variation. Two metal mesh sheets with different densities are also sourced in this test.
2 PRAY
. 144 .
1 SPRAY
DENSE MESH
3 SPRAY
LOOSE MESH
. 145 .
Test 05.4: Architectural Elements Affecting Facial Recognition
TRANSLUCENCY TEST
0 mm
5 mm
10 mm
20 mm CLEAR
. 146 .
1 SPRAY
2 SPRAY
0 mm
5 mm
10 mm
20 mm 3 SPRAY
LOOSE MESH
DENSE MESH
. 147 .
Test 06.4: Architectural Elements Affecting Facial Recognition
TRANSLUCENCY TEST
. 148 .
Closer the material is to the camera, blurrier the face is, and therefore the recognition is more like to fail. The blurrness can affect the accuracy of the recognition. In the case of the metal mesh, the recognition became unstable, not exactly relating to the distance, but more about the angle the mesh is positioned. Face was more recognisable with certain patterns the mesh creates over the face. Material with different translucency can be strategically deployed to manipulate view range and create privacy in public space.
. 149 .
Test 05.5: Architectural Elements Affecting Facial Recognition
REFLECTION
. 150 .
/// TEST SETUP
. 151 .
Test 05.5: Architectural Elements Affecting Facial Recognition
REFLECTION TEST
. 152 .
Face can certainly be recognised through mirror with its reflection captured by the camera. Mirror expands the CCTV's view range in space. It can be strategically use to cover blink spots which not directly captured by CCTV.
. 153 .
. 154 .
Chapter 6 Contextualisation 6.1 TEST 06: Sculpting Space - View Cones Mapping - Activities Capturing - CCTV's View
TEST 06: SCULPTING SPACE
. 157 .
Test 06: Sculpting Space
SPECIFY CAMERA
. 158 .
. 159 .
Test 06: Sculpting Space
SIMULATE CAMERA Input Focal Length and Sensor Size
Calculates Field of View (View Cone) Horizontal 99.27o Vertical 66.99o
. 160 .
Test 06: Sculpting Space
SIMULATE CAMERA
99.27o
1984 pixels 66.99o 1105 pixels
20 m
View Cone Horizontal 99.27o Vertical 66.99o Vision 20m 1984 (H) * 1105 (V) Pixels
. 161 .
Test 06: Sculpting Space
VIEW CONE GEOMETRY
View Cone Horizontal 99.27o Vertical 66.99o Vision 20m 1984 (H) * 1105 (V) Pixels
. 162 .
Test 06: Sculpting Space
VIEW CONE GEOMETRY RENDER
View Cone Horizontal 99.27o Vertical 66.99o Vision 20m 1984 (H) * 1105 (V) Pixels
. 163 .
Test 06: Sculpting Space
RECONSTRUCT SPACE
. 164 .
Test 06: Sculpting Space
OVERLAY VIEW CONE ONTO SPCAE
CCTV 1
CCTV 3 CCTV4
CCTV 2 CCTV5
. 165 .
Test 06: Sculpting Space
FACIAL RECOGNITION RANGE
. 166 .
r
r r
CCTV view cones overlaid on the restaurant model visualised the threshold of space being captured. From previous test, it was evident that in order to deply facial recognition, the camera setup need to zoom into the subjects rather than using wide lens captureing large space. In this case, the camera has a fixed lens, so if facial recognition is applied to the camera, it would only success within certain distance r.
. 167 .
Test 06: Sculpting Space
CAPTURING ACTIVITY AND FACE 1
2 3 1
2 3 1
2 3 1
2 3 1
2 3 . 168 .
Test 06: Sculpting Space
CAPTURING ACTIVITY AND FACE
Body activity captured by CCTV 1, 2 &3 Face captured by CCTV 1 & 2
Face is positioned within reasonable distance and perspective in CCTV 1&2's view range, facial recognition may be successful.
Body activity captured by CCTV 1, 2 &3 Face captured by CCTV 3
Face is positioned within reasonable distance and perspective in CCTV 3's view range, facial recognition may be successful.
Body Activity captured by CCTV 1, 2 &3 Face Captured by CCTV3
Body Activity captured by CCTV 1, 2 &3 Due to the perspective of the face captured by CCTV 3, facial recognition might fail. Face Captured by CCTV3
Body Activity captured by CCTV 1, 2 &3 Due to the perspective of the face captured by CCTV 3, facial recognition might fail. Due Face Captured by CCTV 1, 2 &3 to the distance between CCTV 1&2 and face, facial recognition might fail.
Body Activity captured by CCTV 1, 2 &3 Due to the perspective of the face captured by CCTV 3, facial recognition might fail. Due Face Captured by CCTV 1, 2 &3 to the distance between CCTV 1&2 and face, facial recognition might fail. . 169 .
Test 06: Sculpting Space
FACIAL RECOGNITION RANGE
. 170 .
This test analysed the movement of the recorded subject. With different positions and gestures, subjects' activity and facial information is recorded or recognised by different cameras located in this restaurant. The test demonstrated that the space can be sculpted from CCTV's view and human trajectory can relate to the way one navigates through view cones.
. 171 .
. 172 .
Chapter 7 Threshold 7.1 New Cartography of the City 7.2 Shaping Behaviours with Shades 7.3 Design from Behaviours 7.4 Conclusion: Threshold
What Can the City See?
A NEW CARTOGRAPHY OF THE CITY
The study developed a graphic language to visualise view cones of CCTV cameras with technical parameters information. Applying this method to a large city scale and combining information of the owners of the CCTV cameras would develop a new cartography of the city, which demonstrate what the city can see through CCTV system. "How consensual is the city?" is asking how surveilled is the city. Visualising view cones is clarifying where consent is needed for practising surveillance in the urban context. As an embodiment of social contract, this cartography shows different social tribes inside the city with territory defined by CCTV view cones. This map give people an agency to choose to give their consent of being watched or not. It uses virtual boundaries to sculpt spaces, and therefore, influences citizens' behaviours and trajectory in the city.
. 175 .
View Cones in City Scale
Taking a section of Notting Hill, this map contains both public and private CCTV locations.
. 176 .
Mapping out Spaces Visible by CCTVs
Interpreting from the map on the left side, this map demonstrated area visible by CCTVs.
. 177 .
Euston Station CCTV View Ranges with Crowds Dynamic Simulation
Restaurant's View Cones and Customer Activity Study from Chapter 6
. 178 .
Territory Defined by Social Contract
SHAPING BEHAVIOURS WITH SHADES
The visualisation of the view cones embodies the virtual boundaries defined by optical devices. This division intensifies the fundamental ethical issue of surveillance, which is not asking people's consent of being watched. By defining the watched territory and the unwatched territory, the map is actually creating two sets of belief systems as a guideline for citizens traveling inside the city. When people waking inside the view cones, they are giving consent to the watcher who owner the CCTV and entering the zone where social contract applies to the specific communities. People who choose not to step into the view cones is believing their own control impact over the institutions, but by not accepting the social contract, they lose the privilege of being part of the community or using public services.
. 179 .
Façades Generated from Line Drawings (Isola et al., 2018)
Generated Dish Images from Recipe (Bar El, Licht and Yosephian, 2019)
. 180 .
Speculation: Surveillance and Machine Learning as a Design Tool
DESIGN FROM BEHAVIOURS
Facial recognition technology has facilitated the existing CCTV surveillance system an efficient processing capacity to identify and locate targeted individuals. In the architectural design realm, this capacity can be deployed to collect user informations to analyse how space is used. Furthermore, with the development of AI image generation (Generative Adversarial Network), spatial images can be generated from a large image database. This is then developed to Image to Image Conditional Adversarial Network as image shown on the left.(https://arxiv.org/pdf/1611.07004.pdf ) Building facades can be generated from sketch line drawings. Another example is the dishes images generated from recipe given in text. Speculating from these technologies, we can expect the further architectural design to take use of all these advanced tools - collect user behaviours information and translate it into customer orientated spatial design.
. 181 .
. 182 .
Conclusion
THE THRESHOLD
The essential contention of contemporary CCTV surveillance and facial recognition is not about potential power abuse since it is not the problem of the system itself but the problem of the authority and the institution owning the system. The fundamental issue is that the surveillance is practising without citizens' consent. It is firstly not asking people's consent and secondly not giving people choices of not giving consent - public areas and services will be inaccessible if one chooses not to be watched by CCTVs. Therefore, asking how consensual is the city is actually asking how surveilled is the city. In this study, "how surveilled" is broken down into detailed technical parameters tested and simulated with hardware and software to explore the threshold of the "watching" territory. The study is ultimately about defining the threshold which can sculpt spaces and influence people's behaviours. The design thesis of this new cartography of the city is using threshold to define two territories representing two different value systems in urban context. Given the spatial experience and functions have always been divided by thresholds (rigid and soft), any architectural design is also about defining thresholds and using thresholds to organising spaces.
. 183 .
. 184 .
Chapter 8 Bibliography
BIBLIOGRAPHY
GitHub. (2019). ageitgey/face_recognition. [online] Available at: https://github.com/ ageitgey/face_recognition [Accessed 13 Apr. 2019]. AL-Allaf, O. (2014). REVIEW OF FACE DETECTION SYSTEMS BASED ARTIFICIAL NEURAL NETWORKS ALGORITHMS. [online] Arxiv.org. Available at: https://arxiv.org/ftp/arxiv/ papers/1404/1404.1292.pdf [Accessed 13 Apr. 2019]. Bar El, O., Licht, O. and Yosephian, N. (2019). GILT: Generating Images from Long Text. [online] Arxiv.org. Available at: https://arxiv.org/pdf/1901.02404.pdf [Accessed 13 Apr. 2019]. 2020 Vision Systems. (2019). CCTV Command & Control | 2020 Vision Systems. [online] Available at: https://www.2020cctv.com/security-solutions/cctv-command-control/ [Accessed 13 Apr. 2019]. Wh-tech.com. (2010). Development of IP Surveillance, Network cameras | WH Inc.. [online] Available at: http://www.wh-tech.com/products/about_camera/development_surveillance.htm [Accessed 13 Apr. 2019]. Draper, R. (2018). They Are Watching You—and Everything Else on the Planet. [online] Nationalgeographic.com. Available at: https://www.nationalgeographic.com/magazine/2018/02/surveillance-watching-you/ [Accessed 13 Apr. 2019]. Eugdpr.org. (2018). EUGDPR – Information Portal. [online] Available at: https://eugdpr.org/ [Accessed 13 Apr. 2019]. Nec.com. (2019). Face Recognition. [online] Available at: https://www.nec.com/en/global/solutions/safety/face_recognition/index.html [Accessed 13 Apr. 2019]. Solid.inrupt.com. (2019). How It Works | Solid. [online] Available at: https://solid.inrupt.com/ how-it-works [Accessed 13 Apr. 2019]. Isola, P., Zhu, J., Zhou, T. and Efros, A. (2018). Image-to-Image Translation with Conditional Adversarial Networks. [online] Arxiv.org. Available at: https://arxiv.org/pdf/1611.07004.pdf [Accessed 13 Apr. 2019]. Mingle, K. (2016). Project Cybersyn - 99% Invisible. [online] 99% Invisible. Available at: https://99percentinvisible.org/episode/project-cybersyn/ [Accessed 13 Apr. 2019]. Statewatch.org. (2007). NATIONAL CCTV STRATEGY. [online] Available at: https://www.statewatch.org/news/2007/nov/uk-national-cctv-strategy.pdf [Accessed 13 Apr. 2019]. Nguyen, A., Dosovitskiy, A., Yosinski, J., Brox, T. and Clune, J. (2019). Synthesizing the preferred inputs for neurons in neural networks via deep generator networks. [online] Papers.nips.cc. Available at: https://papers.nips.cc/paper/6519-synthesizing-the-preferred-inputs-for-neurons-in-neural-networks-via-deep-generator-networks.pdf [Accessed 13 Apr. 2019]. Technomate. (2019). TM-102 HD EF. [online] Available at: http://www.technomate.com/products/TM%252d102-HD-EF.html [Accessed 13 Apr. 2019].
. 187 .
NOTTING HILL CARNIVAL OPERATION BREAKDOWN