Ayodhya: How this startup will assist UP Police with AI to boost security for Ram Mandir inauguration

We had already developed a software called Trinetra for UP Police in 2018 which digitised all the criminal records with the police. We connected the same database (of 800,000 people with criminal backgrounds) to live cameras in Ayodhya. So, all the suspicious people and criminals recorded in the past are now being monitored using Jarvis.

The AI technology can also do a reverse facial recognition. Suppose the police has some intelligence of a particular person who might cause trouble. All they have to do is upload a picture of the person on Jarvis and the software will search in the camera’s history for any previous videos.

The cameras also have Automatic Number Plate Recognition (ANPR) capabilities. Through access to government’s vehicle registration database including e-vahan Parivahan, they can identify vehicles with fake number plates.

The technology also enables the surveillance cameras to perform attribute-based searches, such as identifying a person from a crowd with specific clothing, colour or style of accessories, accompanying child, etc. And it’s not just restricted to humans. The software can even look for a specific car with a signage or mark on it.

What is the specification of the camera and what distances does it record?

We are using 400 cameras with 4 mega pixels that can record objects and humans up to a distance of 15 feet very clearly. These are in the yellow zone — which include the temple entry gate and surrounding hot spots, such as Kanak Bhawan, Hanuman Garhi, Shri Nageshwar Nath Mandir, Ram Ki Paidi, etc. It’s almost the entire city of Ayodhya.

Did you do a recce to prepare for January 22?

We have had several meetings at the UP police headquarters. Our team has been in Ayodhya since December 20 to test the placement of the camera, height, orientation, review footage and check software.

How accurate is this technology?

The technology is benchmarked on LFW (Labeled Faces in the Wild — a database of face photographs designed for studying the problem of unconstrained face recognition) and NIST (The National Institute of Standards and Technology) databases which is the global standard for measuring accuracy. On those databases, our accuracy is 99.7 per cent. We have published over 25 research papers and have received two patents for our tech. We have further benchmarked our audio-based identification technology on VoxCeleb database (on 1 million samples) and achieved accuracy of 98.5 per cent.

How many people are expected for the inauguration and what’s the kind of security threat you are anticipating and preparing for?

We are anticipating 20,000-50,000 people after January 22 in Ayodhya. But on that day, there will be 15,000 people (8,000 people in the guestlist and the remaining 7,000 will constitute security personnel, workers and others).

The threat perception is always high at an event of this scale and importance. The main priority is to monitor the crowd and look out for people with past criminal records and suspicious vehicles.

Have you handled an event of this magnitude before?

We have worked previously on the G20 summit and continue to work with nine state police departments offering AI-enabled analytics to track criminals, review security breaches, violence, overcrowding and unauthorised access. We have been monitoring 3,000 cameras in UP prisons. Around 30,000-plus criminals and 17 terrorist modules have been busted using this technology.