When Covid forced everyone to hide behind masks, one lot was happy: the criminals. Now they could find it easier to disappear into the crowd as identifying a person became more difficult.
With such a threat, there was talk of the deployment of a facial recognition system that can detect faces with masks or disguises during the pandemic.
The National Crime Records Bureau, in fact, had floated a tender for it last year, but no announcement was made on whether this technology was developed or deployed.
However, it turns out that Defence Research and Development Organisation (DRDO), India's premier defence laboratory, has indeed developed such a system.
Dubbed 'Face Recognition System under Disguise', or FRSD, the system claims to detect faces through several "disguises like face masks, beard, moustache, wigs, sunglasses, head-scarves, monkey-caps, hats, etc".
The Ministry of Defence (MoD) recently released a report called ‘AI in Defence’, which revealed FRSD and other three facial recognition systems developed by organisations under MoD for the Indian Army.
Since these technologies may not just be reserved for military activities, but also be deployed in public places, it is necessary to throw light on how and why they are being used.
Instead of human eyes, the FRSD relies on algorithms to identify the person from patchy, low-resolution surveillance camera feeds.
“The algorithm can also be used by security agencies for robust face search across large repositories,” the MoD report said.
The system can be deployed in restricted/ secure zones for live video surveillance. It can also be deployed in public places to recognise anti-social elements, the report said.
It takes into consideration different lighting conditions, shadows on the face, crowd occlusions and so on for identification.
“‘Face recognition in the wild’ on surveillance camera feeds is a difficult problem to solve due to the low resolution of the images captured from the cameras. This problem becomes even more challenging to solve with the added complexity of various facial disguises, crowd occlusions and varied illuminations,” the MoD report said.
DRDO has developed the system keeping in mind that it should be scalable across servers and graphic processing units.
“The system comes with a flexible video analytics suite with a number of additional surveillance applications like people counting, geo-fencing, fire detection and collision detection.”
Project Seeker is a facial recognition system developed by entities under the MoD.
Developed and deployed by the Indian Army, it has been designed for population monitoring, surveillance and garrison security, according to the MoD report.
It doesn't require internet connectivity, can accrue intelligence data from multiple sources and be set up remotely with a field-ready system anywhere.
It can be deployed in ‘disturbed’ areas for continuous surveillance and monitoring, as well as at civilian establishments ‘for ensuring state-of-the-art security’.
“The Seeker system is a self-contained, AI-based facial recognition, surveillance, monitoring, and analysis system for identification & tracking of threats for counter-terrorism, continuous surveillance, and monitoring of disturbed areas,” the report said.
It said the system can be deployed in ‘critical military’ or ‘civilian establishments’ for added security.
Using intelligence data from various sources, the Army aims to track the movement of terrorists and ‘anti-national’ elements.
The Army aims to achieve "psychological dominance on threats and anti-national elements", the report said, while explaining how the technology will serve the nation.
It is important to note that there is no legal definition for the term ‘anti-national’, and has not been defined in Statutes.
Robot at the border
Apart from Project Seeker, the Indian Army has also developed Silent Sentry, which is a fully, facial recognition capable, 3D-printed rail-mounted robot that slides on a rail and can be installed on fences and anti-filtration obstacle system (AIOS).
The robot which communicates through WiFi is embedded with artificial intelligence for detecting human beings and faces.
“The video feed received from the robot is analysed by an AI software utilising object recognition. The software detects movement and human presence automatically, generates an audio alarm and stores the photographs with time and date log,” the report said.
On detection of a human, a background facial recognition algorithm is activated, which tries to determine the identity of a person from a stored database. The facial feature information is then stored in the database.
Driver fatigue monitoring system
BEML Ltd, a public sector company under the MoD, has developed a driver fatigue monitoring system which uses facial recognition.
“Assessing driver fatigue in critical conditions is an indispensable tool, especially in the Armed Forces,” the report said.
The report said that the system detects the onset of drowsiness in a driver while the vehicle is in motion.
A camera inside the cabin films the driver continuously, and an algorithm analyses the footage frame by frame and determines whether the driver’s eyes are open or closed.
“Detection is done by continuously looking out for symptoms of drowsiness, while considering physical cues including yawning, drooping eyelids, closed eyes and increased blink durations by using the percentage of eyelid closure over the pupil over time (PERCLOS) algorithm,” it added.
As dazzling as these technologies may sound, at the end of the day, these are all based on algorithms and the software that is being deployed.
So how reliable are these systems, given that they are documented to be prone to error?
There are concerns over mis-identification due to poor accuracy in correctly identifying faces.
“Facial recognition technology is inaccurate. It throws up faulty results. And now with masks, which can cover half the face, the accuracy will go even lower,” said Anushka Jain, associate counsel at Internet Freedom Foundation (IFF).
For instance, in a test conducted in 2018, Amazon’s facial recognition tech known as Rekognition incorrectly matched 28 members of US Congress, identifying them as other people who have been arrested for a crime.
Jain gave the example of two siblings who can be wrongly identified while wearing masks.
“Two siblings wearing masks can very well have similar looking upper-half of the face. They can be wrongly identified. This can even lead to communities being targeted,” Jain said.
The report does not mention the accuracy of the FRSD technology. Moneycontrol has reached out to the DRDO for comments in this regard, and the copy will be updated when a response is received.
"Any decision taken on account of any misinformation might lead to dire consequences. Facial recognition, as a practice, in its application, can also have shortcomings of its own. Therefore, the data thus received needs to be subject to a process where the Armed Forces need to further scan and filter the retrieved data,” Kritika Seth, Founding Partner at Victoriam Legalis – Advocates & Solicitors.
Over the years, the deployment of facial recognition technology by state governments and the Centre for governance and policing has been under the scanner of civil society groups and digital rights activists, who worry about privacy infringement.
Although the usage of the system will be more concentrated on foreigners, Seth from Victoriam Legalis raised queries on the data collection practices, and on whether its usage aligns with the Right to Privacy judgement.
“There is no legal framework which mandates transparency in data collection for the above purpose. The opacity regarding the use of personal data can be a violation of the right to privacy as given in the case of Justice K.S. Puttaswamy (Retd.) versus Union of India,” Seth said.
“Furthermore, the Army is also keen to monitor social media pages. Such surveillance will intersect with already existing state surveillance and might not fall within the purview of roles of Armed Forces,” she added.
Moneycontrol has reached out to the Indian Army with queries in this regard, and the story will be updated when a response is received.
Siddharth Suresh, partner at DSK Legal, explains that data collected from facial recognition solutions comes within the purview of “Biometric Data” and is classified as “sensitive personal data under the Information Technology (Reasonable security practices and procedures and sensitive personal data of information) Rules, 2011 (“SPDI Rules”).
However, he said, regulations have carved out exemptions for government agencies to collect and use such data without the consent of the data subject, with the underlying presumption being that such use of data is for the general public good and national security.The recent notification of the Criminal Procedure (Identification) Act, 2022 also allows authorities to collect and share biometric information.”