In crime fight, AI cameras give Bengaluru cops a leg-up

Bengaluru: Facial recognition software being used by the Bengaluru police has identified 2.5 lakh faces with a criminal background across the city in the last 90 days, leading to the arrest of at least 10 people, according to data accessed by DH.

The Facial Recognition System (FRS) was introduced on a pilot basis in Bengaluru under the Safe City Project. The cameras are connected to the FRS set up at the police command centre. 

Police are now focusing on stabilising and improving the system's accuracy. 

"The ability of the system to accurately match a subject’s facial features with the criminal database needs to be stabilised. We are currently testing two facial recognition software at the command centre, and will adopt the best one,” Police Commissioner B Dayananda told DH. 

He added that the goal is to bring the FRS to the same level of accuracy as the Automatic Number Plate Recognition (ANPR) cameras.

As part of the Safe City project, 7,500 AI-powered cameras were installed to recognise faces. Of these, police purchased licences for 1,000 cameras to link with the FRS. 

An officer at the command centre noted the system has achieved a 75-80% accuracy, which is being increased to 98-99 per cent through machine learning and data reading. "Based on the stabilisation results, we will consider acquiring more licences," he said. 

The system has come in handy for the police in some cases.

For example, in August, a Tumakuru police constable came to Bengaluru in search of a habitual offender involved in 40 cases. The FRS helped pinpoint the suspect's location, leading to his capture near Sadashivanagar after a dramatic chase, said an officer. 

The FRS was also used to create the route map of a techie who recently went missing from Bengaluru and was found in Noida with a changed appearance, the officer said. He gave this example to stress that the system is also effective in cases of missing persons and abduction. 

Another police officer told DH that the system is very effective in crowded places and large gatherings, such as political gatherings and festival celebrations. This allows the police to monitor live feeds and closely track any individuals with criminal backgrounds present in the crowd, noted another officer.

Highlights - How FRS works Cameras are linked to facial recognition software set up in the command centre Software is fed with criminal database maintained by the Karnataka police Camera sends its feed to FRS-connected monitors in the command centre If any person with a criminal background enters the range of any camera, it sends a notification to the command centre If a wanted person comes in the range of the camera, it sends an alarm to the command centre 

Quote - The ability of the system to accurately match a subject’s facial features with the criminal database needs to be stabilised. We are currently testing two facial recognition software at the command centre and will adopt the best one - B Dayananda Bengaluru Police Commissioner