Chennai’s T. Nagar, arguably India’s biggest shopping district by revenues and crowded on any given day, gets even more packed in festival seasons as thousands throng its saree and jewellery stores. Every year, Deepavali, less than three months away this year, presents the perfect hunting ground for pickpockets and other petty thieves — and a headache for the local police.

This time, however, the city police have reason to believe it has a handle on things. It has a technology that analyses CCTV footage to spot, in real time, people with a criminal history visiting the T. Nagar area. “We are matching real-time CCTV video footage with our criminal database using the FaceTagr system and if any criminals are identified in that area, we get an immediate alert and we can further investigate,” says P Aravindan, deputy commissioner of police. Last year, FaceTagr, a face recognition software developed by an eponymous Chennai company, was used in a few areas with results that convinced the police to spread it to all of the T Nagar area, he adds.

Aravindan’s counterparts in Punjab are as big fans of real-time surveillance as him. Amritsar Police used something the state’s police calls Punjab Artificial Intelligence System, or PAIS, developed by Gurugram AI company Staqu Technologies, to solve a murder case within 24 hours — again, using CCTV footage and facial recognition technology. The company has piloted a camera mounted on a pair of smart glasses to capture a real-time feed and analyse it for facial matches with a database.

Elsewhere, the Surat Police has a picture intelligence unit that relies on NEC’s proprietary NeoFace technology for facial recognition, as also vehicle number plate recognition, to track persons of interest. The result is alerts that the police can proactively act upon and faster turnaround in solving cases. Surat can claim to be a step ahead of Tokyo: NEC plans to use the latest version of its NeoFace technology at the 2020 Tokyo Olympics to track accredited persons – athletes, officials, media, and others – at multiple venues.

Welcome to the Big Eye helping law keepers and administrators in India to instantly recognise faces and use the information in multiple use cases.

Facial recognition and image cognition tech is nothing new, to be sure. We have seen them in movies for some time now – be it the Jason Bourne series in which the CIA uses complex surveillance tech to track the agent or the Mission Impossible movies where the protagonist use facial recognition to get access to secure areas. Or, the recent Steven Spielberg movie, Ready Player One, in which the villain uses camera drones. This kind of advanced – and even futuristic – image recognition-based surveillance all set to go mainstream in India with the rapid proliferation of cameras: from the public and private CCTVs to the ubiquitous mobile phone cameras.

Investigation on steroids

Chennai-based FaceTagr has been working with Indian Railways since last year to prevent human trafficking. “Finding missing children and the prevention of human trafficking was one of the first use cases that we developed. We work with the Indian Railways, state police departments, and CBI to prevent human trafficking,” says Vijay Gnanadesikan, CEO and co-founder, FaceTagr.

His moment of epiphany that led to the idea for developing FaceTagr was on a morning drive to work in Chennai traffic and watching children begging at his window. “I reached the office and discussed with my cofounder. We realised that there is an existing database of missing children with photographs and, with face recognition technology, we could develop a solution that could help solve the problem and in a way also prevent human trafficking,” says Gnanadesikan. Cut to today: the tool has been deployed at the India-Nepal and India-Bangladesh borders at nearly 24 checkpoints to monitor human trafficking.

FaceTagr is a face recognition technology that works on both static images and video footage. The same technology is being used in a solution for the Chennai police to identify criminals. “Earlier a suspect had to be taken to the police station, fingerprinted, and then his details were verified. Imagine a guy walking on the road at 2 am who is looking suspicious. A police patrol can take the suspect’s photograph with our app and, within a second, receive details about his crime history,” says Gnanadesikan.

The T. Nagar deployment runs on real-time CCTV footage. In the areas it was deployed last year, the system helped reduce the number of crimes “from three digits to a single digit” during last year’s Deepavali season, claims the FaceTagr CEO.

The system compares the real-time CCTV footage of the crowd with the police criminal database for facial matches. “Once someone from the database is identified among the crowd, the picture shows up, which is then re-verified by the police personnel monitoring the system for a reconfirmation,” says Gnanadesikan, adding that an ID match does not mean a crime is committed. “Someone might also be there for shopping and we and the police team are very mindful of that, but it will give the police a notification about the person’s whereabouts in the area.”

One of the clever outcomes of the deployment is that the system helps identify criminals from other cities or areas. According to DCP Aravindan, a police officer in Chennai city will likely not know of a criminal from, say, Tirunelveli, Kanyakumari or other far off places. This is where the face recognition system comes in handy, he says.

“Traditionally, we have data of all criminals station-wise and there is also a crime team which is familiar with the criminals and can recognise them. But, of late, with the improvement in connectivity and communication, people from far-off places come and commit a crime and this has made it challenging to identify them,” he says. The state’s crime database currently has over 60,000 photographs with more photographs being added daily. Every week, the department nabs two or three criminals with the help of the face recognition system, Aravindan adds.

Are there any privacy concerns? “To avoid misuse we have conducted multiple training programs for all the police personnel who are using this application and we have instructed them that unless they find a person suspicious, they should not take a photograph. We have designed an SOP (standard operating procedure) for using the system to avoid misuse,” adds the deputy commissioner.

Surveillance on smart glass

The face recognition system of Staqu, the Gurgaon AI startup, has been deployed in the states of Uttarakhand, Punjab and Rajasthan.

According to Atul Rai, Staqu’s CEO and co-founder, different law enforcement jurisdictions or agencies, even within a state, often have their own sets of data and it becomes difficult to sift through them and find links or patterns. Staqu’s answer to that problem was ABHED, short for Artificial Intelligence Based Human Efface Detection, which formed the base software for a mobile application and is connected to a backend database processing system. “This system accumulates images, speech and text, and using all this information, it develops intelligence for these agencies,” says Rai.

The company has also developed a real-time video surveillance-based face recognition technology that works via a camera mounted on a smart glass. The system was piloted with the Punjab Police and the company is now in the process of deploying with the Dubai Police, says Rai.

Most CCTVs today have a limited view and, in comparison, an officer wearing the smart glass and moving in a crowd will have a better field of view, says Rai. “In real time, the glass will stream the video footage to the server, which will then match the footage and give the report if any person from the database is detected,” he adds.

The Staqu-developed PAIS, or Punjab Artificial Intelligence System, can image match with an accuracy of 98% if the database has five images of the person, claims Rai.

Another use case for face recognition technology that has been coming up in India is in the corporate sector for attendance and security.

“In many of the enterprise use cases, the technology is used in controlled spaces – for example, conferences where most attendees pre-register or employees access systems in companies,” says Uday Chinta, managing director of American technology service company IPSoft, which has also developed and deployed an AI-based personal assistant called Amelia in the US. “Amelia is able to recognise a person using his facial features and able to assist them and give personalised service based on their identity,” says Chinta.

Software services company Tech Mahindra has launched a facial recognition system for employee attendance at its Noida office. According to one report, the system also comes with a “moodometer” that will track the mood and emotions of employees and give additional analytics to the company.

Beyond face analytics, image recognition technology is also being used to identify vehicles. The National Highways Authority of India has been using AI-based image recognition systems to tag and identify vehicles across its infrastructure in the country.

Underlying digital layer: databases

The scarier part to the tech is its dark side: mass surveillance covering all. Countries like China have already deployed mass surveillance on its citizens. Chinese citizens today have a scoring system assigned to them by the government based on various factors including data captured through the surveillance program which will give the preferential access to services like fast internet access.

In the case of India, to facilitate proper surveillance in a state, one of the first requirements is a digital database which already exists in many forms across central and state governments. With or without a double take, the answer is obvious: Aadhaar, India’s citizen ID database. With a population of 135 crore and Aadhaar covering over 90% of this population, it is India’s most extensive database.

Notwithstanding the use cases detailed earlier in this story and the huge interest among state police and law enforcement agencies in India, collecting data and using it – even it is to bust crime – falls into grey areas. In June this year, news reports had National Crime Records Bureau director Ish Kumar saying that investigators need to be given limited access to Aadhaar. Reacting to this, the Unique Identification Authority of India (UIDAI) issued a statement saying that access to Aadhaar biometric data for criminal investigation is not permissible under Section 29 of the Aadhaar Act, 2016 — which perhaps explains why the Punjab Police declined requests for interviews for this story.

Longtime Aadhaar critic Sunil Abraham, executive director of Bengaluru’s Centre for Internet and Society (CIS), calls Aadhaar “the perfect tool for surveillance”.

“The main database is the Aadhaar database. It’s got your iris and biometrics information already and they have said that they will strengthen the fingerprint authentication with facial recognition. So now, they have the have the full surveillance infrastructure that they need. The collection devices (CCTVs) are just there to collect the data but the actual recognition engine is Aadhaar only,” says Abraham, who is leaving CIS to join non-profit Mozilla Foundation as a vice president in January.

According to him, all three types of biometrics – fingerprint data, iris information data, and facial data – can be used in a remote and covert fashion and, therefore, in a non-consensual fashion. (Editor’s note: There is no public incident, to date, that proves such a use.)

Abraham is “100% sure” where we are headed. “The reason why I call Aadhaar a surveillance project is not that there is metadata stored, I call it a surveillance project because the biometrics are being stored. Metadata is one of the problems, that is the profiling risk but the surveillance risk primarily comes from the biometric data that they have,” he says. By metadata, he is referring to a citizen’s information such as phone number, age, sex, address, and other details.

There are also other databases in the works that could provide the basis for surveillance. Like: the Crime and Criminal Tracking Network & Systems (CCTNS) across police stations in India. According to the CCTNS website, as of May 2018, the CCTNS hardware and software deployment has covered nearly 94% of the police stations across India. There have been reports of the CCTNS system being used as a mass surveillance system in the guise of e-policing by authorities in Hyderabad.

Early in 2016, the Hyderabad of Police had launched a tender looking for companies to set up a citizen profiling and monitoring system. According to a report in Telangana Today, the Integrated People Information Hub (IPIH) gives the police access to personal informations of its citizens including names, family details, addresses and other related information by sourcing them from documents like police records, FIRs and other external sources like utility connections, tax payments, voter identification, passport etc.

During Israeli Prime Minister Benjamin Netanyahu’s visit to India in January, Tel Aviv-based AI company Cortica had announced a partnership with India’s Best Group to develop solutions for combing through data captured daily by drones, surveillance cameras, and satellites. The aim is to develop an AI-based real-time identification of patterns, concepts and situational anomalies to identify potential problems, flag them and improve safety in the process. More details such as scale and scope of this partnership are not available at this point in time.

Mass surveillance: Easier said than done

Take a step back. India already has multiple digital surveillance – even if not mass, real-time facial recognition – programs in place to keep track of its citizens. E.g.: the Central Monitoring System or CMS and NATGRID, which is short for National Intelligence Grid. Others include Telecom Enforcement Resource and Monitoring or TERM and NETRA (NEtwork TRaffic Analysis) surveillance software developed by the Centre for Artificial Intelligence and Robotics. These are just some of the surveillance programs operated by governments of the day, both central and state.

But when it comes to mass surveillance in real time, even with the AI-based tech is available today, the currently installed infrastructure might not be ready for real-time mass surveillance. “Countries like China are good at setting up infrastructure which is very essential for mass surveillance systems to be in place,” says Kedar Kulkarni of Bengaluru-based deep learning startup Hyperverge, who also insists that all CCTVs out there today might not be fit to conduct facial recognition.

According to Kulkarni, for a mass surveillance system to be in place, you either need cameras that can capture and do computing for face recognition within its hardware or you need a robust network which can transmit live feeds from multiple cameras to processing centres, which is very bandwidth intensive.

Most public spaces in India including railway stations, bus depots, metro station, marketplaces are often under CCTV surveillance. New Delhi is all set to have one of the largest deployments in the country of CCTVs with the state government announcing plans to install 1.4 lakh CCTVs across Delhi. The India Railways is also setting aside Rs 3,000 crore in its 2018-19 budget to install CCTV systems across 11,000 trains and 8,500 stations, according to a news report.

In comparison, China is said to have 170 million CCTV cameras installed across the country currently and this number is estimated to go up by 400 million in the next three years, says a BBC news report.

Even the staunchest privacy activists acknowledge what surveillance can deliver if used carefully. “Overall, it is a very powerful technology. It should be used for law enforcement, it should be used for national security. That is the correct domain of application,” says Abraham. He hastens to add the caveats: “When we use it, we have to use it with lots of safeguards and it should be used only on a very small subset of the population. It shouldn’t be a technology that is broadly deployed in the population because it is not necessary, it is not proportionate, and the risks are very high.”

The flip and funny side of facial recognition-based surveillance is that the government does not need the technology to actually work. Just the threat of surveillance – that big brother is watching you – is enough to reduce crime. According to Gnanadesikan, the Chennai CEO of FaceTagr, one reason for the drop in crime rate in last year’s T. Nagar trials was that criminals knew that they were being watched.

---